EJSCREEN
Environmental Justice
Mapping and Screening Tool
EJSCREEN
Technical Documentation
May 2015
-------
EJSCREEN
Environmental Justice Mapping and Screening Tool
EJSCREEN Technical Documentation
May 2015
U.S. Environmental Protection Agency
Office of Policy
Washington, D.C. 20460
Suggested citation:
U.S. Environmental Protection Agency (EPA), 2015. EJSCREEN Technical Documentation.
For more information:
www.epa.gov/ejscreen
-------
ACKNOWLEDGEMENTS
This document was prepared by the United States Environmental Protection Agency's Office of Policy
(OP) with drafting, editing and technical support from Abt Associates. ESRI and SAIC, working with the
Office of Environmental Information (OEI), obtained and processed datasets obtained from the Census
Bureau, US Department of Transportation (DOT), and various EPA offices including the Office of Air
(OAR), Office of Research and Development (ORD), Office of Water (OW), and Office of Solid Waste and
Emergency Response (OSWER). Ongoing input, review, and outreach have been provided by all of these
offices as well as EPA's Regional offices, Office of Chemical Safety and Pollution Prevention (OCSPP),
Office of International and Tribal Affairs (OITA), Office of Enforcement and Compliance Assurance
(OECA), Office of Environmental Justice (OEJ), and Office of General Counsel (OGC). External expert peer
reviewers provided comments on the methods and documentation as well. Specific acknowledgement is
given to the contributions of Mark Corrales, Bridgid Curry, William Nickerson, and Fred Talcott in the
design of EJSCREEN and the preparation of this document.
-------
Contents
1 INTRODUCTION 6
Environmental Justice at EPA 6
Environmental Justice Mapping and Screening at EPA 7
Development of EJSCREEEN 8
Purposes and Uses of EJSCREEN 8
Caveats and Limitations of EJSCREEN 9
Advances in EJ Screening Provided by EJSCREEN 11
2 OVERVIEW OF DATA AND METHODS IN EJSCREEN 13
Environmental Indicators Selected for EJSCREEN 13
Demographics in EJSCREEN 16
Using Demographics as Proxies for Potential Susceptibility 16
Demographic Indicators Included in EJSCREEN 20
Demographic Indexes in EJSCREEN 21
Environmental Justice Indexes in EJSCREEN 22
How the EJ Index Works 22
Supplementary EJ Indexes 24
EJ Indexes, Population Density, and Rural Areas 25
Why the EJ Indexes are Not all Combined 26
PERCENTILES 27
What a Percentile Means 27
Color-coded High Percentile Bins 28
Buffer Reports 30
What the Buffer Report Calculates 30
Choosing a Buffer Size versus Rationale for Distance in Proximity Indicators 30
3 DETAILS ON THE ENVIRONMENTAL INDICATORS IN EJSCREEN 33
Environmental Factors Not Included 33
Environmental Factors in EJSCREEN 35
4 BIBLIOGRAPHY 67
Appendices
Appendix A. Development of EJSCREEN 78
Appendix B. Technical details on percentiles, rounding, buffering, and demographic data 80
Appendix C. Technical details on proximity indicators 100
Appendix D. Summary statistics for indicators 107
Appendix E. Formulas for demographics and EJ indexes 115
Appendix F. Quality control / quality assurance 118
Appendix G. Peer review 120
Appendix H. Initial filter approach for screening 122
iv | P a g e
-------
Tables
Table 1. Types of Environmental Indicators Included in EJSCREEN 15
Table 2. Summary Table of Environmental Indicators and Sources 35
Table 3. Likelihood of Lead-Based Paint Hazards by Housing Construction Date 49
Table 4. Tallies of 2008-12 ACS Block Groups Used in 2015 Version of EJSCREEN 85
Table 5. ACS Tables Underlying EJSCREEN Demographic Data and Lead Paint Indicator 87
Table 6. Summary Statistics for Environmental Indicators 108
Table 7. Summary Statistics for Demographics 109
Table 8. Spearman Correlation Coefficients for Environmental Indicators Ill
Table 9. Spearman Correlation Coefficients for Demographic Indicators 112
Figures
Figure 1. Histograms of Block Group Environmental Indicators as ratio to mean value (log scale shows mean
value as zero) 113
v | Page
-------
Introduction
EJSCREEN Environmental Justice
Mapping and Screening Tool
Technical Documentation
1 INTRODUCTION
The United States Environmental Protection Agency (EPA) is charged with protecting human health and
the environment for all Americans. In order to better meet the Agency's responsibilities related to the
protection of public health and the environment, EPA has developed a new environmental justice (EJ)
screening tool, called EJSCREEN. In some ways, EJSCREEN is similar to prior screening or mapping tools.
As a new tool, however, it offers improvements such as easy web-based access to powerful mapping
and data reporting tools, a wide range of updated demographic information, environmental indicators
addressing more topics, and higher resolution maps covering the entire nation. EJSCREEN also provides
standard reports that bring together environmental and demographic data in the form of EJ indexes.
These are summarized as percentiles to put the information in perspective and facilitate comparisons
between locations.
EJ screening tools may be used to explore one location using a data report, or to look across a wide area
using maps. EJ tools have been used in a wide variety of circumstances, and EJSCREEN can support a
similarly broad range of applications. EJSCREEN provides useful data and indicators, and highlights
places that may be candidates for further review, including additional consideration, analysis or
outreach.
This document describes EJSCREEN within the context of EPA's EJ program, and provides details on the
data and methods used to create the indicators and indexes in EJSCREEN. The Appendices in this
document provide additional detail on data and methods for interested users.
Environmental Justice at EPA
Since EJ mapping and screening is just one aspect of EPA's ongoing commitment to environmental
justice, it is helpful to understand the broad, historical context of EPA's EJ work.
EPA has defined "environmental justice" as follows:
Environmental Justice is the fair treatment and meaningful involvement of all people regardless
of race, color, national origin, or income with respect to the development, implementation, and
enforcement of environmental laws, regulations, and policies.... Fair treatment means that no
group of people should bear a disproportionate share of the negative environmental
6 | P a g e
-------
Introduction
consequences resulting from industrial, governmental and commercial operations or policies.1
(italics added)
EPA's efforts to understand EJ concerns date back at least to EPA's 1992 report on Environmental Equity
(U.S. EPA, 1992). The 1992 report documented health and exposure disparities associated with
race/ethnicity and income. To address such disparities, in 1994, Executive Order 12898 (EO 12898)
mandated that each covered federal agency make achieving environmental justice part of its mission by
identifying and addressing, as appropriate, any disproportionately high and adverse human health or
environmental effects of its programs, policies and activities on minority, low-income, tribal and
indigenous populations.2
These early activities provided a foundation for EPA's continued commitment to environmental justice,
which was reaffirmed in January 2010 when EPA Administrator Lisa P. Jackson announced Expanding the
Conversation on Environmentalism and Working for Environmental Justice as one of the Agency's top
seven priorities. EPA has made great progress implementing this priority and has worked across the
Agency to make a difference in overburdened communities, a goal that has been reiterated by EPA
Administrator Gina McCarthy.
EPA has been engaged in a variety of research and analytic efforts related to environmental justice, to
support both regulatory analysis and screening efforts. Recent efforts have been guided by Plan EJ 2014,
which was released for public comment in July 2010 and finalized in September 2011.3 EPA's Office of
Environmental Justice (OEJ), in conjunction with the rest of the Agency, works to protect human health
and the environment in communities overburdened by environmental pollution by integrating
environmental justice into all EPA programs, policies and activities.
Environmental Justice Mapping and Screening at EPA
Mapping tools as well as screening-level applications have a substantial history at EPA, and EJ is one area
in which maps and screening can be useful. Several EPA Regional offices have used basic screening tools
that map demographic information and allow staff to overlay selected environmental data such as
facility locations. In connection with EJSEAT, a screening tool that was developed by EPA's Office of
Enforcement and Compliance Assurance (OECA), the National Environmental Justice Advisory Council
(NEJAC) provided recommendations to EPA about how to design an EJ screening tool in its May 2010
report, "Nationally Consistent Environmental Justice Screening Approaches" (NEJAC, 2010).
These various early screening tools have been used for internal EPA purposes only, and generally were
not available to the public. EPA's main publicly available EJ mapping tool until 2015 was EJVIEW,4 a web-
based tool that displayed selected demographic and environmental data, and allowed users to overlay
these data on maps of a community or wider area.
1 http://www.epa.gov/environmentaliustice/basics/index.html, accessed 10/16/2014.
2 http://www.archives.gov/federal-register/executive-orders/pdf/12898.pdf
3 http://www.epa.gov/environmentaliustice/plan-ei/index.html
4 http://www.epa.gov/compliance/ei/mapping.html
7 | P a g e
-------
Introduction
Development of EJSCREEEN
Plan EJ 2014 included a commitment to develop a nationally consistent environmental justice screening
and mapping tool in order to better meet the Agency's responsibilities related to the protection of
public health and the environment in a manner that is consistent with EO 12898 and the goals of Plan EJ
2014. This commitment was the impetus behind the development of EJSCREEN. This new effort
provided an opportunity to reassess and build upon prior efforts, while considering new data, new
scientific findings, new analytic methods and a variety of policy considerations.
The goal in developing EJSCREEN has been to take account of this prior progress and learning, and
provide a new, user-friendly screening tool that addresses policy questions and stakeholder concerns in
an informative manner. An important part of this effort has been to ensure the screening tool reflects an
appropriate balance between simple, feasible, screening-level information on the one hand, and high-
quality data and strong science on the other.
Development of EJSCREEN began in late 2010, and EPA staff began using an early version in 2012, as
prior tools such as EJSEAT were phased out. EJSCREEN was peer reviewed in early 2014 through a letter
review (see Appendix G), and updated with newer data and an improved interface in 2014/2015.
EJSCREEN was released to the public in 2015, replacing EJVIEW as EPA's public-facing EJ mapping tool.
Purposes and Uses of EJSCREEN
EJ mapping and screening tools combine environmental and demographic indicators in maps and
reports. This information can help to highlight geographic areas and the extent to which they may be
candidates for further review, including additional consideration, analysis or outreach. The tools also
allow users to explore locations at a detailed geographic level, across broad areas or across the entire
nation. Environmental indicators typically are direct or proxy estimates of risk, pollution levels or
potential exposure (e.g., due to nearby facilities). Demographic indicators are often used as proxies for a
community's health status and potential susceptibility to pollution. Environmental and demographic
data and indicators may be viewed separately or in combination.
This type of screening information may be of interest to communities as well as many other
stakeholders, and also can support a wide range of research and policy goals. In general, EPA's efforts
are more effective and efficient if they are informed by an understanding of where the impacts of
existing pollution may be greatest. Screening tools can also help ensure that such areas are not
overlooked, and receive appropriate consideration, analysis or outreach.
Screening tools can be appropriately put toward a wide variety of uses. The public has used EJVIEW in a
variety of ways, and is likely to use EJSCREEN in many ways as well. EPA has used existing internal EJ
screening tools in aspects of enforcement, compliance, the Superfund program, permitting, and
voluntary programs. Screening tools also have been used in developing retrospective reports, and to
enhance geographically based initiatives. EJSCREEN will be able to support a similarly wide variety of
uses.
8 | P a g e
-------
Introduction
Screening tools should be used for a "screening-level" look. Screening is a useful first step in
understanding or highlighting locations that may be candidates for further review. However, it is
essential to remember that screening-level results do not provide a complete assessment of risk, and
have significant limitations.
Caveats and Limitations of EJSCREEN
EJSCREEN is a pre-decisional screening tool, and was not designed to be the basis for agency decision-
making or determinations regarding the existence or absence of EJ concerns. It also should not be used
to identify or label an area as an "EJ Community." Instead, EJSCREEN is designed as a starting point, to
highlight the extent to which certain locations may be candidates for further review or outreach.
EJSCREEN's initial results should be supplemented with additional information and local knowledge
whenever appropriate, for a more complete picture of a location. Additional considerations and data,
such as national, regional, or local information and concerns, along with appropriate analysis, should
form the basis for any decisions.
EJSCREEN, as a screening tool, is more limited than a detailed analysis in two key ways. First, it has data
on only some of the relevant issues, and second, there is uncertainty in the data it does have. It is
important to understand each of these limitations.
The first limitation arises because a screening tool cannot capture all the relevant issues that should be
considered (e.g., other local environmental concerns). Any national screening tool must balance a desire
for data quality and national coverage against the goal of including as many important environmental
factors as feasible given resource constraints. Many environmental concerns are not yet included in
comprehensive, nationwide databases. For example, data on environmental factors such as local
drinking water quality and indoor air quality were not available with adequate quality, coverage and/or
resolution to be included in this national screening tool. EJSCREEN cannot provide data on every
environmental impact and demographic factor that may be important to any location.
The second important limitation is that EJSCREEN relies on demographic and environmental estimates
that involve substantial uncertainty. This is especially true when looking at a small geographic area, such
as a single Census block group. A single block group is often small and has uncertain estimates. A buffer
that is roughly the same size as a block group or smaller will introduce additional uncertainty because it
has to approximate the locations of residences. Therefore, it is typically very useful to summarize
EJSCREEN data for a larger area, covering several block groups, in what is called a "buffer" report, as
explained later in this document. There is a tradeoff between resolution and precision: Detailed maps at
high resolution can suggest the presence of a local "hotpot," but are uncertain. Estimates based on
larger areas will provide more confidence and precision, but may overlook local "hotpots" if not
supplemented with detailed maps.
The demographic uncertainty combined with uncertainty in environmental data means EJ index values
are often quite uncertain for a single block group. Therefore, modest differences in percentile scores
between block groups or small buffers should not be interpreted as meaningful because of the
uncertainties in demographic and environmental data at the block group level. We do not have a high
9 | P a g e
-------
Introduction
degree of confidence when comparing or ranking places with only modest differences in estimated
percentile. For this reason, it is critical that EJSCREEN results be interpreted carefully, particularly for
individual block groups, and that additional information be used to supplement or follow up on
screening, where appropriate.
The demographic estimates, such as percent low-income, come from the American Community Survey
(ACS) from the United States Census Bureau. The ACS is comprised of surveys, not a full census of all
households. This means the Census Bureau may estimate that a block group is 30% low-income, for
example, but it might actually be 20% or 40% in some cases (see Appendix B for a discussion of
uncertainty in demographics).
Uncertainties are also discussed in section 2 (with regard to buffer reports), and Appendix B (in
discussions of buffering details and demographic data).
Related to the issue of uncertainty is that fact that the environmental indicators are only screening-level
proxies for actual exposures or health risks. This is particularly true for the proximity indicators, for
example. Even for the indicators that directly estimate risks or hazards, as with the air toxics cancer risk
indicator, estimates have substantial uncertainty because emissions, ambient levels in the air, exposure
of individuals, and toxicity are uncertain. Section 3 provides technical details on each environmental
indicator.
The inclusion of a dataset in EJSCREEN does not imply it is the newest, best, or primary estimate of
actual conditions or risks. Estimates are based on historical data and may not reflect current or future
conditions. The vintage of environmental indicators varies and is not the same as the vintage of the
demographic data. The NATA air toxics indicators and the PM2.5 and ozone indicators in particular should
be viewed with this in mind, because emissions related to PM2.5, ozone, and air toxics generally have
decreased in recent years. This version of EJSCREEN incorporated the most recent data that were
available at the time of each indicator's development. Every attempt will be made to use the most
recent appropriate data available in future updates of EJSCREEN. There is always a delay between the
release of raw data and their eventual incorporation into any models, tools, or maps. It is also useful to
note that although the raw numbers for some indicators do not represent current conditions, the
percentiles are much more likely to be reasonably representative of today's conditions in most
locations. This is because even if emissions have been significantly reduced overall, for example, the
differences between various locations are unlikely to have changed as dramatically, especially when the
reductions have come from national regulations and other trends affecting entire industries or sectors in
many locations. For this reason, the percentiles may be more representative of current conditions than
the raw values of the indicators. Finally, some supplementary maps and local information can
complement the EJSCREEN indicators to provide more recent information. In particular, EJSCREEN also
provides updated maps of PM2.5 and ozone nonattainment areas (areas not meeting national ambient
air quality standards).
There are also some limitations in geographic coverage - EJSCREEN lacks data in some locations for
some indicators, such as in Alaska, Hawaii, and Puerto Rico.
10 | Page
-------
Introduction
In short, as with any screening tool, the indicators in EJSCREEN cannot address all the considerations
that may be relevant to a given situation, they are often only a screening-level proxy for a given issue,
and in any case, there is significant uncertainty involved, particularly for a single block group. For these
types of reasons, among others, it generally is not appropriate to rely on any screening tool as the basis
for a key decision. It is often very useful to obtain information on other issues not included in EJSCREEN,
updated information when available, as well as local knowledge, data, and concerns.
Advances in EJ Screening Provided by EJSCREEN
EJSCREEN offers a variety of enhancements relative to previous approaches to EJ mapping and screening
or analysis. For example, the tool includes updated demographic information (from the ACS) rather than
relying on the Census that is conducted every ten years. It also provides several new environmental
indicators, covering a wider range of issues such as traffic volume and proximity. EJSCREEN includes a
suite of EJ indexes that quantify the combination of environmental and demographic indicators. It
includes high resolution maps, and a new geospatial software and data system that improves access to
new tools with a simple, browser-based interface and centralized, consistent data. The buffer reports
are calculated using detailed Census block data for more accurate estimates of where residents are
located. EJSCREEN provides access to a great deal of data, and presents standardized reports. These
reports use summary metrics and percentiles to facilitate national, regional or state-level perspectives
and a better understanding of EJ issues.
EJSCREEN can help explore the environmental, demographic and EJ characteristics of a block group or
buffer area. It provides numerical estimates for each place, for both environmental and demographic
data, such as the traffic proximity indicator, or the percentage of local residents who are racial/ethnic
minorities.
EJSCREEN also presents multiple "EJ Indexes" for each place. An EJ Index is a way of combining, in a
quantitative way rather than only visually on a map, the environmental indicator and the demographic
information for a location. A separate EJ Index is provided for each environmental indicator in EJSCREEN,
for each block group in the US. The EJ Index goes beyond a simple visual overlay of maps of environment
and demographics, to actually quantify the extent to which these two factors co-occur.
In EJSCREEN, the basic level of geographic resolution is the Census block group. Each block group is
defined by the U.S. Census Bureau, with a logical and unambiguous numbering scheme, and associated
digital shape files that permit mapping with modern geographical information system (GIS) software.
Block group data are widely used by researchers and others. Block groups also provide a relatively stable
framework; for instance, block groups are not subject to frequent boundary definition changes that
political jurisdictions and postal ZIP codes may experience.
Estimates in EJSCREEN are compiled by block group, and that is the most detailed level at which results
can be viewed. However, demographic estimates for a single block group are often based on a small
sample of the local population, and are uncertain. Similarly, some environmental indicator estimates
are derived from lower-resolution data, and all involve uncertainty. Therefore, it is typically very useful
and advisable to summarize EJSCREEN data within a larger area that covers several block groups, in what
11 | Page
-------
Introduction
is called a "buffer" report. An EJSCREEN user can specify or draw buffers of custom sizes and shapes as
needed. For example, a buffer could include all residents within 1 mile of a certain location. When a
buffer covers several block groups, it provides an estimate that has less uncertainty than a single block
group or smaller buffer would. EJSCREEN summarizes data for all residents within some distance from a
selected point, using a circular buffer, or within a user-defined buffer of any shape, using Census blocks
(not just block groups) to refine estimates of how many residents are inside the buffer, as explained in
Appendix B.
12 | Page
-------
Overview of Data and Methods
2 OVERVIEW OF DATA AND METHODS IN EJSCREEN
This section describes the environmental and demographic data used in the tool, as well as the methods
used to combine them and produce EJ Indexes. As of 2015, EJSCREEN contains 12 environmental
indicators, which range from estimates of human health risk to proxies for potential exposure such as
proximity to hazardous waste sites. The tool also contains six demographic indicators, which are
combined into two separate demographic indexes (the demographic index uses the average of two
indicators, and a supplementary demographic index uses the average of all six). A demographic index is
combined with an environmental indicator to create an associated EJ Index. The environmental,
demographic, and EJ indicators and indexes are all calculated for each block group, and can be
summarized within a defined buffer area. The sections immediately below summarize the
environmental data, demographic data, and EJ indexes. Section 3 provides more detail on each
environmental indicator.
Environmental Indicators Selected for EJSCREEN
Some environmental indicators used in EJSCREEN quantify proximity to and the numbers of certain
types of potential sources of exposure to environmental pollutants, such as nearby hazardous waste
sites or traffic. The lead paint indicator indicates the presence of older housing, which often, but not
always, indicates the presence of lead paint, and therefore the possibility of exposure. In some cases,
the term "exposure" is used very broadly here to refer to the potential for exposure. Others indicators in
EJSCREEN are estimates of ambient levels of air pollutants, such as PM2.5, ozone, and diesel particulate
matter. Still others are actual estimates of air toxics-related cancer risk, or a hazard index, which
summarizes the ratios of ambient air toxics levels to health-based reference concentrations. In other
words, these environmental indicators vary widely in what they indicate, as discussed further below.
A variety of considerations has informed the selection of these environmental indicators; in general, the
selected indicators exhibit the following characteristics:
Resolution: Screening level data are available (or could be readily developed) at the block
group level (or at least close to this resolution).
Coverage: Screening level data are available (or could be readily developed) for the entire
United States (or with nearly complete coverage).
Relevance to EJ: Pollutants or impacts are relevant to EJ (e.g., differences between groups
have been indicated in exposures, susceptibility, or health endpoints associated with the
exposures)
Public health significance: Pollutants or impacts are potentially important in the United
States (e.g., notable impacts estimated or significant concerns have been expressed, at least
locally, or exposure has been linked to health endpoints with substantial impacts
nationwide).
EPA selected environmental indicators after a review of data availability (including the criteria and
review of data availability for an Environmental Quality Index, now provided by county in Messer, Jagai,
13 | P a g e
-------
Overview of Data and Methods
Rappazzo, & Lobdell (2014), and described in Lobdell, Jagai, Rappazzo, & Messer (2011)); health
disparity information (e.g., CDC's major 2011 report on health disparities (Centers for Disease Control
and Prevention, 2011a)); risk ranking studies (e.g., Unfinished Business (U.S. EPA, 1987), Reducing Risk
(U.S. EPA, 1992), and related reports); and risk estimates from major studies (e.g., related to PM2.5
ambient standards). EPA also reviewed data from the CDC (Centers for Disease Control and Prevention,
2011c) and other sources in the federal government (Fedstats, 2007), and consulted EPA Regions and
program offices that are responsible for data collection and analysis under EPA's key environmental
statutes. The State of California's work on CalEnviroScreen5 was also tracked throughout its
development. Other internal EPA data tools were also examined, including EJSEAT, EJVIEW, C-FERST, and
tools used in EPA Regional Offices.
After review, EPA selected the following environmental factors for use in the first version of EJSCREEN:
Air pollution:
o PM2.5 level in air.
o Ozone level in air.
o NATA air toxics:
¦ Diesel particulate matter level in air.
¦ Air toxics cancer risk.
¦ Air toxics respiratory hazard index.
¦ Air toxics neurological hazard index.
Traffic proximity and volume: Amount of vehicular traffic nearby, and distance from roads.
Lead paint indicator: Percentage of housing units built before 1960, as an indicator of
potential exposure to lead.
Proximity to waste and hazardous chemical facilities or sites: Number of significant
industrial facilities and/or hazardous waste sites nearby, and distance from those:
o National Priorities List (NPL) sites,
o Risk Management Plan (RMP) Facilities.
o Hazardous waste Treatment, Storage and Disposal Facilities (TSDFs).
o National Pollutant Discharge Elimination System (NPDES) permitted major direct
dischargers to water.
Each of these environmental indicators is explained in detail in section 3, and Appendix D provides
summary statistics for the indicators. Again, it is important to understand that these indicators vary in
how relevant they are to actual estimated risks to health or welfare, and how significant those impacts
may be. These indicators represent a spectrum in terms of the quality of information about potential
impacts, ranging from direct estimates of risk to rough indicators of proximity or exposure to pollution
or other environmental hazards. Table 1 provides more detail on how closely each environmental
indicator in EJSCREEN approximates actual estimated risk.
5 http://oehha.ca.gov/ei/ces2.html at http://oehha.ca.gov/ei
14 | Page
-------
Overview of Data and Methods
Table 1. Types of Environmental Indicators Included in EJSCREEN
Indicator
Place on Exposure-
Risk Continuum
Key
Medium
NATA Air Toxics Cancer Risk
Lifetime inhalation cancer risk
NATA Respiratory Hazard Index
Ratio of exposure concentration to RfC
Risk/Hazard
NATA Neurological Hazard Index
Ratio of exposure concentration to RfC
NATA Diesel PM (DPM)
(Hg/m3)
Air
Particulate Matter (PM2.5)
Annual average (ng/m3)
Ozone
Summer seasonal average of daily maximum 8-hour
concentration in air (ppb)
Potential Exposure
Lead Paint Indicator
Percentage of housing units built before 1960
Dust/ Lead
Paint
Traffic Proximity and Volume
Count of vehicles (average annual daily traffic) at major roads
within 500 meters, divided by distance in kilometers (km)
Air/ Other
Proximity to RMP Sites
Count of facilities within 5 km, divided by distance
Proximity to TSDFs
Count of major TSDFs within 5 km, divided by distance
Proximity/ Quantity
Waste/
Water/ Air
Proximity to NPL Sites
Count of proposed and listed NPL sites within 5 km, divided by
distance6
Proximity to Major Direct Water Dischargers
Count of NPDES major facilities within 5 km, divided by distance
Water
Abbreviations:
NATA
National Air Toxics Assessment
RfC
Reference concentration from EPA's
NPL
National Priorities List, Superfund program
Integrated Risk Information System
NPDES
National Pollutant Discharge Elimination System
PM2.5
Particulate matter (PM) composed of
RMP
Risk Management Plan
particles smaller than 2.5 microns
TSDFs
Hazardous waste Treatment, Storage, and
Mg/m3
micrograms of PM2.5 per cubic meter of air
Disposal Facilities
ppb
parts per billion, of ozone in air
6 Count of NPL sites excludes deleted sites and sites in U.S. territories.
15 | P a g e
-------
Overview of Data and Methods
It is also important to note that each proximity indicator focuses on one category of facility or site (e.g.,
NPL), but the category's facilities or sites vary in the degree to which they could actually pose risks. They
vary in the amount of emissions (if any), the possibility of exposure to any pollutants released, the size
of the facility or site, and toxicity of the pollutants or severity of the impacts that might occur. As a
screening tool, EJSCREEN generally does not distinguish based on these factors in proximity indicators
(although NATA indicators do account for such information). Any closer review of a particular location
would have to consider these important differences.
All of these indicators are focused on potential impact at residential locations (e.g., proximity of
residence to traffic), and therefore only address some of the exposures that individuals may face. Data
are generally insufficient to readily estimate exposures away from the home, particularly in a screening
tool. Exposures that occur away from the home, such as at work, at school or during a commute, are not
captured in EJSCREEN unless those exposures are near the home or in other locations that happen to
have the same level of exposure.7
Demographics in EJSCREEN
This section describes why demographic indicators are included in EJSCREEN, which specific
demographic indicators were selected, and what data are used to derive the demographic indicators.
Using Demographics as Proxies for Potential Susceptibility
EJSCREEN has been designed in the context of Executive Order 12898s which ordered the following:
To the greatest extent practicable and permitted by law, and consistent with the principles set
forth in the report on the National Performance Review, each Federal agency shall make
achieving environmental justice part of its mission by identifying and addressing, as appropriate,
disproportionately high and adverse human health or environmental effects of its programs,
policies, and activities on minority populations and low-income populations9
EJSCREEN was also designed in the context of EPA's EJ policies, including EPA's Interim Guidance on
Considering Environmental Justice During the Development of an Action (U.S. EPA, 2010). That guidance
document explained EPA's focus on demographics as an indicator of potential susceptibility or
vulnerability to environmental pollution:
To help achieve EPA's goal for EJ (i.e., the fair treatment and meaningful involvement of all
people), EPA places particular emphasis on the public health of and environmental conditions
affecting minority, low-income, and indigenous populations. In recognizing that these
populations frequently bear a disproportionate burden of environmental harms and risks ... EPA
works to protect them from adverse public health and environmental effects of its programs.
7 A partial exception is the data from NATA, which make some attempt to include some nonresidential exposures,
as explained in NATA's technical documentation (http://epa.gov/nata/).
8 http://www2.epa.gov/laws-regulations/summarv-executive-order-12898-federal-actions-address-environmental-
justice
9 http://www.archives.gov/federal-register/executive-orders/pdf/12898.pdf
16 | Page
-------
Overview of Data and Methods
EPA should pay particular attention to the vulnerabilities of these populations because they have
historically been exposed to a combination of physical, chemical, biological, social, and cultural
factors that have imposed greater environmental burdens on them than those imposed on the
general population. (U.S. EPA, 2010, p. 4)
EJSCREEN uses demographic indicators as very general indicators of a community's potential
susceptibility to the types of environmental exposures included in this screening tool. Impacts of
pollutants depend on a combination of exposure and susceptibility to those exposures. Demographic
factors may be related to both of these. Therefore, it is very useful to distinguish between 1) the fact
that some demographics are associated with higher exposure, and 2) the fact that demographics are
useful in predicting susceptibility to those exposures. To indicate potential exposures, EJSCREEN uses
environmental indicators, not demographics. EJSCREEN uses demographics to indicate potential
susceptibility. EJSCREEN then combines the exposure and susceptibility indicators in the form of an EJ
Index.
The demographic indicators in EJSCREEN are a way to indicate which communities may be more
susceptible to a given level of exposure to environmental pollutants. For example, individuals may be
more susceptible when they are already in poor health, have reduced access to care, lack resources or
language skills or education that would help them avoid exposures or obtain treatment, or are at
susceptible life stages. Nationwide direct measures of health status are not available for all block groups
or even tracts - such data are typically compiled by county in national databases. Demographics,
however, are available for every block group, and are correlated with health status and these other
susceptibility factors, making them useful screening-level indicators of potential susceptibility at the
local level.
Note that this report uses the term susceptibility in a qualitative, general sense, to refer to what various
authors have called susceptibility and/or vulnerability. Susceptibility in this report means greater
"impact" for a given environmental indicator value. The terms vulnerability and susceptibility sometimes
are used interchangeably, although various other reports and programs have made distinctions between
these terms.10
The relationships between demographics, exposure, and susceptibility are complex. For example,
demographics may be associated with susceptibility to pollutants in any of the following ways:
Greater personal exposure despite the same ambient level of pollutant. For example,
children have higher breathing rates or ingest more lead dust than adults (U.S. EPA, 2011a),
10 For example, EPA's 2009 National Ambient Air Quality Standards (NAAQS) documents (U.S. EPA, 2009b) and also
EPA's Regional Vulnerability Assessment program treat susceptibility and vulnerability as essentially identical
(http://www.epa. gov/reva /glossa rv. htm I). Other EPA definitions have addressed particular contexts, such as in
EPA glossaries (http://www.epa.gov/OCEPAterms/vterms.html, http://www.epa.gov/OCEPAterms/sterms.html),
and a report on vulnerability to climate change (U.S. EPA, 2009a). One National Academies report (Science and
Decisions) distinguished between the two terms (National Research Council, 2009). In other contexts, the terms
have varied uses - see Villagran de Leon (2006) for a detailed comparison of various definitions of vulnerability in
the context of natural disasters, or work on vulnerability indexes for developing countries by Briguglio (1997).
17 | Page
-------
Overview of Data and Methods
and certain groups may tend to encounter or be less able to avoid certain exposures due to
limited resources, language barriers, education, cultural practices, or lack of information.
Susceptibility because of a greater percentage increase in health risk for a given exposure,
e.g., "effect modification" or "multiplicative interaction" may occur. An example would be
where cumulative previous exposure means a group is more likely to be closer to a
threshold for adverse effects, or where greater stress/allostatic load increases susceptibility
through inflammatory or other pathways. Several examples of effect modification relevant
to EJ and PM2.5 are referenced by Bell & Ebisu, 2012 and in a review of subgroups
susceptible to ozone (Bell, Zanobetti, & Dominici, 2014). A growing body of research has
documented interactions of psychosocial stress and environmental exposures.
Susceptibility because of higher baseline risk or rates of pre-existing diseases. The same
percent increase in mortality risk has a larger impact on absolute risk if baseline risk is
higher.
Susceptibility because of increased overall burden resulting from an initial health risk (e.g.,
because of less ability to recover due to lack of health care or resources). For example, low-
income or minority individuals, or those with less than a high school education, are far less
likely to have health insurance (Cohen & Martinez, 2011).
One reason for EJSCREEN to focus on potentially susceptible demographic groups is that a large body of
research has documented health disparities between demographic groups in the United States, such as
differences in mortality and morbidity associated with factors that include race/ethnicity, income and
educational attainment (e.g., Centers for Disease Control and Prevention, 2011a; Galea, Tracy, &
Hoggatt, 2011). For example:
About two thirds (65%) of non-Hispanic white adults reported excellent or very good health
in 2009. In contrast, less than half (49%) of non-Hispanic black adults and 52% of Hispanic
adults reported excellent or very good health.11
Residents with lower income report fewer average healthy days than others (Centers for
Disease Control and Prevention, 2011a), and report worse health overall (Centers for
Disease Control and Prevention, 2010).
Both lower income and minority race/ethnicity have independent associations with higher
asthma rates, particularly among children, and diabetes rates differ greatly by race/ethnicity
(Centers for Disease Control and Prevention, 2011a).
Mortality rates for cancer and heart disease vary somewhat by race/ethnicity (Centers for
Disease Control and Prevention, 2011b). Coronary heart disease and stroke are elevated
among black individuals but generally not in other minority subgroups (Centers for Disease
Control and Prevention, 2011a).
11 This survey, the National Health Interview Survey, represents the U.S. non-institutionalized civilian population,
and presents age-adjusted estimates based on household interviews (Centers for Disease Control and Prevention,
2011d).
18 | Page
-------
Overview of Data and Methods
Infant mortality is higher among non-Hispanic black, American Indian/Alaska Native, and
Puerto Rican (but not other Hispanic) populations (Centers for Disease Control and
Prevention, 2011a).
While some health disparities are due to differences in health care, diet, activities, psychosocial stress or
even genetics, it is possible that some portion of certain disparities may be related to differences in
environmental exposures. Some of these differences in exposure are associated with residential
location, and could be considered in EJSCREEN (while others cannot be considered in EJSCREEN, such as
those related to use of consumer products or diet, for which high-resolution geographic data are not
available). Various environmental exposures have been shown to vary by race/ethnicity, income and
other demographic factors (Liu, 2001; Maantay, Chakraborty, & Brender, 2010; U.S. EPA, 2006a), but
EJSCREEN is not predicated on an assumption of such a correlation.
In addition to Executive Order 12898's call, perhaps the most important reason to focus on key
demographic groups in EJSCREEN is that a growing body of research has shown that demographic
factors are associated with susceptibility - certain groups are more impacted by a given level of
exposure to certain pollutants. Various groups have shown increased susceptibility to certain pollutants,
but further evidence is still emerging in this area and data are limited. Evidence currently available
includes the following:
Certain demographic groups, such as those with lower educational attainment, children, the
elderly and those with low socio-economic status (SES), appear to be more susceptible to a
given exposure to particulate matter (U.S. EPA, 2009b).
Blood lead's association with cardiovascular outcomes appears to be stronger among
Mexican Americans and non-Hispanic blacks than non-Hispanic whites (U.S. EPA 2011c).
Some but not all studies suggest lead has a greater impact on IQ among low SES than high
SES individuals (U.S. EPA 2011c).
EJSCREEN is not designed to explore the root causes of differences in exposure. The demographic
factors included in EJSCREEN are not necessarily causes of a given community's increased exposure or
risk. This does not limit their usefulness for the limited purposes of the screening tool, however - these
demographic factors are still useful as indicators of potential susceptibility to the environmental factors
in EJSCREEN. They may be associated with susceptibility, whether or not they are causal, and can be
used as proxies for other harder-to-measure factors that would better describe or determine
susceptibility but for which nationally consistent data are not available. EJSCREEN screens geographic
areas for increased potential for exposure and increased potential for susceptibility to exposures.
Additional analysis is always needed to explore any underlying reasons for differences in susceptibility,
exposure or health.
Some studies have begun to quantify the degree of susceptibility to specific pollutants in particular
demographic groups, such as the work on educational attainment as an effect modifier for risks
associated with PM2.5 exposure, but such emerging knowledge is still limited to a handful of pollutants
19 | Page
-------
Overview of Data and Methods
and demographic factors (U.S. EPA, 2009b). EJSCREEN, as a screening tool, does not attempt to use this
type of emerging quantitative information.
Demographic Indicators Included in EJSCREEN
A wide range of demographic descriptors have been used by researchers and in EJ screening tools to
represent the "social vulnerability" characteristics of a disadvantaged population (for example, see
deFur et al., 2007, and Bell & Ebisu, 2008).
Executive Order (EO) 12898, addressing EJ issues, refers to low-income and minority populations. We
define these two core factors as:
Low-Income: The number or percent of a block group's population in households where the
household income is less than or equal to twice the federal "poverty level."12
Minority: The number or percent of individuals in a block group who list their racial status as
a race other than white alone and/or list their ethnicity as Hispanic or Latino. That is, all
people other than non-Hispanic white-alone individuals. The word "alone" in this case
indicates that the person is of a single race, since multiracial individuals are tabulated in
another category - a non-Hispanic individual who is half white and half American Indian
would be counted as a minority by this definition.13
Based on a review of other factors used in various EPA EJ screening tools, the four other factors most
commonly used by EPA Headquarters and Regions for EJ analyses are also included in EJSCREEN. The
other four factors are:
Less than high school education: The number or percent of people age 25 or older in a block
group whose education is short of a high school diploma.
Linguistic isolation: The number or percent of people in a block group living in linguistically
isolated households. A household in which all members age 14 years and over speak a non-
English language and also speak English less than "very well" (have difficulty with English) is
linguistically isolated.
Individuals under age 5: The number or percent of people in a block group under the age of
5.
12 More precisely, percent low-income is calculated as a percentage of those for whom the poverty ratio was
known, as reported by the Census Bureau, which may be less than the full population in some block groups. More
information on the federally-defined poverty threshold is available at
http://www.census.gov/hhes/www/povertv/methods/definitions.html. See Appendix B for details on using twice
the poverty threshold.
13 Census definitions of race/ethnicity are available at:
http://www.census.gov/population/www/socdemo/race/index.html and the questions asked about race are
available at: http://www.census.gov/acs/www/about the survey/questions and why we ask/
20 | P a g e
-------
Overview of Data and Methods
Individuals over age 64: The number or percent of people in a block group over the age of
64.
The source of all demographic data used in EJSCREEN is the American Community Survey (ACS) five-year
summary file, which the U.S. Census Bureau compiles yearly.
Appendix D provides summary statistics for the demographic indicators.
The supplementary reports and maps provided by EJSCREEN also include an extensive list of additional
demographic variables, including statistics on race/ethnicity subgroups (e.g., percent Hispanic or Latino),
languages spoken (e.g., % speaking Vietnamese), income (% in poverty), and many other factors. This
supplementary information may be very useful. For example, subgroups within the broad category of
"minority" can differ greatly in their baseline health, exposures, geographic locations, and other factors.
Demographic Indexes in EJSCREEN
The Demographic Index in EJSCREEN is created using the two demographic indicators that were
explicitly named in EO 12898, low-income and minority. For each Census block group, these two
indicators are simply averaged together.
Demographic Index = (% minority + % low-income) / 2
A Supplementary Demographic Index is also available in EJSCREEN, and is the average of all six
demographic indicators.14
Supplementary Demographic Index =
(% minority + % low-income + % less than high school education + % linguistic isolation +
% individuals under age 5+ % individuals over age 64) / 6
14 Census evaluates each characteristic for a different population, so the denominators are not the same across
factors. For example, the denominator for "% less than high school education" is the population in the block group
25 years and older. The denominators for the factors "% minority" and "% low-income" are both close to the total
block group population, but some people in each block group are not evaluated for each demographic
characteristic.
21 | P a g e
-------
Overview of Data and Methods
Users can also view each demographic indicator separately in EJSCREEN, in reports or in maps.
The Demographic Indexes count each indicator as adding to overall potential susceptibility of the
population in a block group, and assumes the demographic indicator have equal and additive impacts.
The current lack of available data precludes any attempt to disentangle the different influences of the
individual demographic indicators, or quantify the degree of overlap or potential synergy between them.
The demographic groups in EJSCREEN overlap to some extent, because some individuals are both low-
income and minority, for example. In fact, these indicators are correlated at the block group level,
because minorities are more likely to be low-income than non-minorities. Appendix D has information
on the correlations between these variables. These correlations do not affect the indicator's ability to
account for susceptibility, if the assumption of additive effects on susceptibility is appropriate. As more
data becomes available in the future, some of these complexities can be reexamined.
Additional information about the demographic data used in EJSCREEN is available in Appendix B.
Environmental Justice Indexes in EJSCREEN
The EJ index is a combination of environmental and demographic information. The environmental
portion of the EJ index is drawn directly from the environmental indicators described above, and the
demographic information is also taken from the demographic indicators above.
How the EJ Index Works
To calculate a single EJ Index, EJSCREEN combines a single environmental indicator with demographic
information. It considers the extent to which the local demographics are above the national average. It
does this by looking at the difference between the demographic composition of the block group, as
measured by the Demographic Index, and the national average (which is approximately 35%). It also
considers the population of the block group.
Mathematically, the EJ Index is constructed as the product of three items, multiplied together as
follows:
EJ Index =
(Environmental Indicator)
X (Demographic Index for Block Group -Demographic Index for US)
X (Population count for Block Group)
The demographic portions of the EJ Index can be thought of as the additional number of susceptible
individuals in the block group, beyond what you would expect for a block group with this size total
population.
22 | P a g e
-------
Overview of Data and Methods
"Susceptible" or "potentially susceptible individuals" are used informally in these examples, as a way to
think of the Demographic Index times the population count in a block group, which is essentially the
average of the count of minorities and count of low-income individuals.15 It is easiest to think of the
average of these counts as "the susceptible individuals" in these examples.
The number of potentially susceptible individuals (Demographic Index times population count) of course
is typically less than the actual number who are minority, low-income, or both. The demographic
breakdown is not reported by block group -the ACS does not provide that level of resolution on the
overlaps.16 For example, suppose that in a certain block group of 1000 people, 350 (35%) are minority
and 350 (35%) low-income. There might be 200 (20%) who are low-income but not minority, and 200
(20%) who are minority but not low-income. In that case, there would be 150 (15%) who are both, and
450 (45%) who are neither. Therefore, there actually would be 550 (55%) who were either minority,
low-income, or both. The Demographic Index would use 35% in this case, which falls between the 15%
who were both minority and low-income, and the 55% who were in at least one of these groups. These
detailed numbers cannot be obtained from the ACS by block group. Therefore, to represent both groups
in a simple way, the average is used.
An extreme example shows another situation: Suppose a block group has 1000 people but is 0%
minority and 100% low-income. The demographic index would be 50%, or the equivalent of 500
"potentially susceptible individuals" in this case. The same would be true in a block group that was 100%
minority but 0% low-income - it would treated as having the equivalent of 50% (500) "potentially
susceptible" for the sake of these examples.
The EJ Index uses the concept of "excess risk" by looking at how far above the national average the
block group demographics are. For example, assume a block group with 1000 people in it. In that block
group, one would expect 350 potentially susceptible individuals (1000 people here x US average of 35%).
However, if the Demographic Index for that block group is 75%, well above the US average, then there
are the equivalent of 750 potentially susceptible people in that block group, or 400 more than expected
for a block group with a population of 1000. The EJ Index would be 400 times the environmental
indicator in this case.
This formula for the EJ Index is useful because for each environmental indicator it finds the block groups
that contribute the most toward the national disparity in that environmental indicator. By "disparity" in
this case we mean the difference between the environmental indicator's average value among certain
demographic groups and the average in the US population.
15 To be precise, the percent low-income times population is not always exactly the same as the count of low-
income residents. The percent low-income is calculated as a fraction of those for whom poverty status could be
determined, which is less than the full population in some block groups. For simplicity, these examples omit that
detail.
16 The closest available data would be table B17001 and related tables, which provide tract resolution cross-
tabulations of race/ethnic groups by poverty status, but this is not available for block groups and does not provide
the income to poverty ratio data needed to calculated "low-income" as defined in EJSCREEN.
23 | P a g e
-------
Overview of Data and Methods
Minority and low-income individuals live in older housing more often than the rest of the US population,
for example. The EJ Index for lead paint (pre-1960 housing) tells us how much each block group
contributes toward this "excess population risk" or "excess number" of people in older housing, for
potentially susceptible individuals. "Excess" in this context simply means the number of potentially
susceptible individuals in older housing nationwide is above what it would be if they were in older
housing at the same rate as the rest of the U.S. population. Locally, it also means the number is above
what it would be if the block group had the same demographic percentages as the U.S. overall.
Analysis of the EJSCREEN data for minority, or for low income, individuals (roughly one third of the US
population in either case), shows they have a higher environmental indicator value on average than the
rest of the U.S. population, for 11 of the 12 environmental indicators (ozone is the exception).
Note that the EJ Index raw value itself is not reported in EJSCREEN reports- it is reported in percentile
terms, to make the results easier to interpret. If one is calculating the actual raw values using the
formula, it is clear that the EJ Index value can be a positive or negative number. A positive number
occurs where the local Demographic Index is above the US average, and this means the location adds to
any excess in environmental indicator values among the specified populations (minority and low-
income) nationwide. A negative value occurs where the local Demographic Index is below the US
average, and it means the location offsets the other locations, reducing any excess in nationwide
average environmental indicator values among minority and low-income populations relative to others.
Most EJSCREEN users will not work directly with EJ Index raw values, however, and positive raw values
for an EJ Index will be presented as higher percentiles and negative raw values will appear as lower
percentiles.
Supplementary EJ Indexes
In addition to this EJ Index formula, two other types of EJ Indexes were tested during EJSCREEN's
development, and those supplementary EJ indexes are also available in maps and the database files. The
three approaches, the basic index and two supplementary approaches, are complementary to one
another and provide different perspectives on a given area. The difference between the three EJ indexes
for any single indicator lies in the different ways the demographic portion of the index is developed.
Analyses comparing the three approaches concluded that they differ only moderately, and that the EJ
Index selected by EPA as the focus of EJSCREEN falls squarely between the other two, in terms of which
locations it highlighted. In other words, the other two approaches were at either end of a continuum,
and did not overlap as much, but the EJ Index featured in EJSCREEN overlaps very substantially with
both of the other two. It also has the advantage of having a clear interpretation, analogous to excess
population risk or the local contribution to excess risk nationwide.
The Supplementary EJ Indexes also combine an environmental indicator with demographic information.
In the first Supplementary EJ Index, the demographic information is simply the Demographic Index for
the block group, times the population of the block group. This index simply omits the demographic index
24 | Page
-------
Overview of Data and Methods
for the nation, so it focuses less on the extent to which demographics exceed the national average. The
first Supplementary EJ Index can be expressed mathematically as follows:
Supplementary EJ Index 1 =
(Environmental Indicator)
X (Demographic Index for Block Group)
X (Population count for Block Group)
The second Supplementary EJ Index can be expressed mathematically as follows:
Supplementary EJ Index 2 =
(Environmental Indicator)
X (Demographic Index for Block Group)
The second Supplementary EJ Index combines an environmental indicator with the Demographic Index
for the block group, leaving out demographic index for the nation and also ignoring the block group
population. This index does not give more weight to block groups with very large numbers of residents.
An advantage is that it is simple and treats each block group equally regardless of how many people live
there. A disadvantage is that this effectively gives less weight to a person in a populous block group than
a person in a less populated block group.
The EJSCREEN mapping application also allows users to see each of these EJ indexes calculated using the
Supplementary Demographic Index, which uses all six demographic indicators instead of just two. In
most locations the results do not differ greatly from what is found using the standard EJ Index.
Details on all the available indexes are also included in Appendix E.
EJ Indexes, Population Density, and Rural Areas
It is very important to understand that the population count per block group is not the same as
population density, so the population weighting of the EJ Index has nothing to do with whether a place
is high density, or rural versus urban. In fact, there is almost no correlation between population count
per block group and population density (population count per square mile covered by the block group),
because in low density areas each block group covers a larger area, keeping the population per block
group fairly consistent. This means population weighting in the EJ Index does not emphasize urban or
high density locations - it is neutral with regard to population density or urbanization. Furthermore, the
vast majority of block groups in the US have similar population counts, so the population weighting in
the EJ Index has a strong influence only in a tiny fraction of locations. For example, about 90% of block
groups had a population between 500 and 2500 in 2008-2012, and only about 1% had a population over
4000.
25 | P a g e
-------
Overview of Data and Methods
It is true that many of the EJ Indexes have higher values in urban, high-density areas, but this is true for
the supplementary EJ Indexes as well, and is not the result of population weighting. Differences in
environmental indicator values (and to some extent percentage demographics) are generally the drivers
of higher EJ Indexes in urban or high-density block groups. NATA indicators and the lead paint indicator,
in particular, are strongly correlated with population density, as are the PM2.5 and traffic indicators to
some extent. The proximity indicators are also positively (but weakly) correlated with population
density. In other words, these environmental indicators appear to be lower in rural areas in general, and
combined with some demographic differences, this tends to make the EJ Indexes lower in those areas.
Relative to those factors, the population weighting (or choice of EJ Index formula) has a very small
influence on whether urban or rural areas are highlighted.
Why the EJ Indexes are Not all Combined
For each environmental indicator, one standard EJ index is available in EJSCREEN. At this time, there is
not a single composite EJ index that combines all the environmental factors. Although it would be useful
if a simple metric could summarize all of the information in EJSCREEN as a single number, there is no
widely-accepted, objective way to combine the differing environmental concerns into one number. This
is because of the value judgments and scientific challenges inherent in deciding how much weight or
importance should be given to each of the environmental factors. They are very difficult to compare, in
terms of public health importance, public concerns, and the many other important considerations that
could be weighed. This topic has been covered extensively elsewhere17, but a very brief explanation may
be useful here.
First, a so-called "equal weighting" does not exist, because it would just be an artifact of the units
(scaling) and aggregation method one chose, which would carry implicit value judgments about how to
weight and combine the factors, even if it seemed simple at first glance. Putting equal weight on each
percentile, for example, would implicitly equate very low risks (e.g., air toxics well below health based
reference levels, as with a neurological hazard index well below 1) and much higher risks (e.g., PM2.5
levels well above a health based standard).
Furthermore, while the use of percentiles provides useful perspective by putting the 12 EJ indexes in
common units, it would be a mistake to assume the 80th percentile, for example, has the same
"importance" for one index or indicator as for another. If two indexes are at the same percentile, it
simply means those two scores are equally common (or equally rare) in the United States. It does not
mean the risks are comparable. It is therefore critical when interpreting EJSCREEN percentiles to also
look at the actual raw numbers for the environmental and demographic indicators.
The challenge is compounded by the fact that rankings of block groups using a composite environmental
index would be quite sensitive to the method chosen to combine the environmental indicators, based
on EPA's analysis of the data. This is also the case, albeit to a lesser degree, for any composite EJ index.
This is a result of the environmental indicators not being highly correlated with each other. The locations
17 See, for example, OECD, 2008, or Finkel & Golding, 1994.
26 | P a g e
-------
Overview of Data and Methods
with the highest PM2.5 levels are not usually the same as the ones with the highest NPL proximities, for
example, as suggested in Appendix D. If the NPL indicator is treated as more important, different block
groups would be highlighted than if the PM2.5 indicator were given more weight. Again, it is important to
acknowledge that there is no objective version of "equal weighting."
For these reasons, the environmental indicators are not combined as a single number, and must be
understood individually for a complete picture. However, they can be viewed all at the same time in a
single tabular report, and this facilitates a broad perspective on all the factors at one time. Those using
EJSCREEN and considering aggregating the data as a single summary metric are strongly urged to
carefully consider these pitfalls associated with doing so. A thorough understanding of each indicator
and the ability to view all of them in a report provides a far better picture of the screening results than
any single number or map is capable of.
Percentiles
What a Percentile Means
EJSCREEN puts each indicator or index value in perspective by reporting the value as a percentile. For
example, an area may show 60% of housing was built prior to 1960. It may not be obvious whether this
is a relatively high or low value, compared to the rest of the nation or in the state. Therefore, EJSCREEN
also reports that 60% pre-1960 puts this area at the 80th percentile nationwide. For a place at the 80th
percentile nationwide, that means 20% of the US population has a higher value.
A percentile in EJSCREEN tells us roughly what percent of the US population lives in a block group that
has a lower value (or in some cases, a tied value). This means that 100 minus the percentile tells us
roughly what percent of the US population has a higher value. This is generally a reasonable
interpretation because for most indicators there are not many exact ties between places and not many
places with missing data.
More precisely, the exact percentile for a given raw indicator value is calculated as the number of US
residents of block groups with that value or lower, divided by the total population with known indicator
values. This is typically the same as or almost exactly the same as dividing by the total US population,
but for some indicators some locations do not have an indicator value. For example, the NATA indicators
are missing for only about one twentieth of 1% of the US population in the 2015 version of EJSCREEN.
The calculated percentile would change by much, much less than 1 percentile point if calculated as a
fraction of the total population instead of as a fraction of those with valid indicator values.
All percentiles in EJSCREEN are population percentiles, meaning they describe the distribution of block
group indicator scores across the population. Note that a population percentile may be slightly different
than the unweighted percentile (the percent of block groups, not people, with lower or tied values),
because not all block groups have the same population size. In practice they are very similar because
very few block groups diverge very much from the average in population size.
27 | P a g e
-------
Overview of Data and Methods
Color-coded High Percentile Bins
Locations at least at the 80th percentile but less than the 90th are shown in yellow on EJSCREEN maps,
while those at the 90th percentile but less than 95th percentile are orange on the maps, and those at the
95th percentile or above are shown in red on maps and reports. These colors call attention to certain
locations as a very simple way to communicate relative screening results. There is no official policy
significance assigned to each individual color on the maps, but the choice of these categories or "bins" is
noteworthy because it signifies that certain ranges of percentiles may merit closer attention.
Percentiles at or above the 95th percentile are shown in red on the EJSCREEN standard report. This is a
way to call particular attention to those cases where the value is in the top 5% of the nation (or region
or state). Indicator or index values in the top 5% tend to be much higher than those in the next 5-10%,
so they may merit close attention. This is especially true for the indicators with highly skewed
distributions, such as the traffic proximity indicator (see Appendix D, Table 6 and Figure 1). For example,
block groups in the top 5% (shown in red on maps and reports) have traffic, NPL, and TSDF proximity
indicators on average that are about three times as high as in the next 5% (shown in orange on the
maps). These differences are far less extreme in the cases of PM2.5 and lead paint indicators, which don't
vary as much across block groups. In general, though, indicator or index values above the 95th percentile
represent much higher demographic, environmental, or EJ Index values than those at lower percentiles.
The maps also identify areas in the 90th to 95th percentiles as orange, and those at the 80th to 90th as
yellow. These additional categories highlight larger groups of locations that have indicator or index
values well above the national mean or median for the given indicator or index. The actual values are
lower than those in the top 5%, typically much lower, but they are still in the top 10 to 20% of values for
the US population overall.
A relatively high percentile means the value is relatively uncommon. However, a high percentile is not
necessarily a real concern from a health or legal perspective. To understand the actual health or other
implications of any screening results requires looking at the actual data and the indicator represents,
and also looking at other relevant data if available. Besides the percentile, other important
considerations in interpreting any screening results include the following:
1. whether and to what extent the environmental data shows values above any relevant health-
based or legal threshold,
2. the significance of any such thresholds, or the magnitude and severity of the health or other
impacts of the given environmental concern, nationally or locally, and
3. the degree of any disparity between various groups, in exposures to the relevant environmental
pollutants.
In maps, EJSCREEN focuses on the US percentiles, as a way to visualize all results in common units.
28 | Page
-------
Overview of Data and Methods
The US percentile uses the US population as the basis of comparison. The state or regional percentile
was calculated based on the population in a given state (or DC) or one of EPA's 10 regions.18 The
national or state or regional mean value was calculated as the population weighted average of the block
groups with data for that indicator, within the respective geographic scope.
Note that the US and state percentiles both will rank block groups in exactly the same rank order within
the given state. If the goal is just to rank or compare locations within a single state, it does not matter
whether the US or state percentile is used. The difference between state and US percentiles becomes
apparent mainly in two situations: when comparing places across states, or when comparing results to
some pre-determined, specific reference percentile (e.g., 80th percentile).
The advantage of US percentiles for an EJ Index, for example, is that a higher percentile in place A versus
place B clearly indicates that the combination of the environmental indicator and demographic index is
greater in place A than place B. In a sense, the US percentile indicates how uncommon it is to have such
a high level for an indicator or index.
State or regional percentiles cannot be compared across states or regions as easily. If two places A and
B, in two different states, happen to both be at the 80th percentile for the traffic EJ Index, for example, it
is not clear which actually has the higher index value. It just means that A's index is just as uncommon
within that state as B's is in B's state. However, this may be useful information because an EJSCREEN
user may want to know how high the indicator is relative to the rest of that state.
The state and US percentiles will be very similar if the state and US average indicator values are very
similar. However, if the state average is very low compared to the US, the state percentile shown will be
higher than US percentile shown, for a given raw value of an indicator. If the state average is much
higher than the US average, for an indicator like the traffic indicator, then a traffic score that would
normally be considered fairly high nationwide, such as the 90th percentile in the US, would not be
considered very unusual within that state, so the state percentile would be lower, and might be only
78th percentile, for example. The state percentile being lower than the US percentile does not mean the
indicator value is lower in the given place, it just means the state average is higher than the US average.
Appendix B provides details on how percentiles are calculated, and rounded when displayed.
18 Regions in the 2014 version of EJSCREEN were defined only by State. A small number of block groups on Tribal
Lands along the borders of Nevada and Arizona are actually part of EPA Region 9, even though they are within the
States OR/ID (Region 10), UT/CO (Region 8), or NM (Region 6). EJSCREEN's percentiles and reports processed those
block groups as if they were in the Region that corresponds to their State. Regional and Tribal Lands maps should
be consulted when viewing those locations and reported Regional percentiles should not be used in those
locations.
29 | P a g e
-------
Overview of Data and Methods
Buffer Reports
What the Buffer Report Calculates
EJSCREEN allows a user to define a buffer, such as the circle that includes everything within 1 mile of a
specific point. Non-circular, user-defined shapes also can be defined to represent buffers of any shape.
The summary within a buffer represents the average resident within the buffer. A report summarizes
the demographics of residents within this buffer, as well as the environmental indicators and EJ index
values within the buffer. It also provides an estimate of the total population residing in the buffer.
Note that this means one cannot compare two buffers of very different population counts without
understanding what each set of results represents. Each represents the average person in that buffer. It
does not represent the absolute total amount the buffer contributes to overall disparity in indicator
scores nationwide or statewide. Even if the two sets of scores are identical other than in population
counts, the buffer with a larger population will contribute more to any national or overall disparity in
indicator scores. In general, however, this situation does not tend to arise because most buffers that a
user creates in practice will be at least roughly similar in size and population. Even if they are not, a user
simply needs to acknowledge that some buffers have larger populations, in which case those percentile
results represent more people.
Appendix B provides the details of buffer calculations.
Choosing a Buffer Size versus Rationale for Distance in Proximity Indicators
An EJSCREEN user's choice of distance for a circular buffer is important, and the considerations need to
be understood.
EJSCREEN is not able to report on buffers that are too large (e.g., ten or more miles in radius) due to
computational limits. It also is not able to report on buffers that are too small (i.e., they do not intersect
the internal points of any Census blocks).
In addition to those limits, as a rule of thumb, it is important to know that a buffer that is as small as the
buffer it centers on will result in estimates with substantially higher uncertainty than one which covers
several block groups. A buffer covering five or more block groups, for example, will provide much more
confidence in demographic estimates (because of sampling uncertainty as well as the challenge of
estimating where residents are located within block groups intersected by the edges of the buffer).
Examining patterns the size of a block group or smaller requires using maps of block groups and Census
blocks, rather than attempting to draw buffers in EJSCREEN.
Uncertainties are also discussed in section 1 (as general caveats), and Appendix B (in discussions of
buffering details and demographic data).
30 | P a g e
-------
Overview of Data and Methods
It is important not to confuse two different distances:
1) the distance a user selects for a circular buffer radius (e.g., by default 1 mile, which is 1.6 km),
and
2) the distance EJSCREEN used in proximity score calculations (i.e., 5 km for facilities and sites, or
500 meters for traffic).
These two are very different, as explained here, because proximity scores use a large distance (e.g., 5
km) and inverse distance weighting, while for circular buffers a user may wish to specify a shorter
distance (e.g., 1 mile or 1.6 km, but at least as large as one or more local block groups) because a buffer
report does not use distance weighting.
The buffer analysis provides a summary of the average resident inside the buffer. It gives equal weight
to each resident, regardless of whether they are closer or further from the center of the buffer. There is
no distance weighting in a buffer report.
Proximity scores were created very differently than buffer reports are calculated. The proximity scores
for each block group were calculated for each residential location using distance weighting to give more
weight to closer facilities, sites, or traffic. Because the proximity score uses distance weighting to focus
less on the more distant points, it was designed to look at a large area using a large radius, or distance (5
km for facilities or NPL sites). By distance weighting, the proximity score can examine this large area and
still provide a useful summary of all the facilities or sites in that wide area.
The proximity score for traffic, for example, looks within a search radius of 500 meters (or further if
none is found in that radius). This distance, or scope, was selected to be large enough to capture the
great majority of road segments (with traffic data) that could have a significant impact on the local
residents, balanced against the need to limit the scope due to computational constraints. Within this
relatively wide zone, the closest traffic is given more weight, and the distant traffic given less weight,
through inverse distance weighting. The same approach was used for the facility or NPL site proximity
scores. A distance of 5 km was chosen to capture the great majority of facilities or sites that could have a
significant impact on local residents. The fact that impacts may be very small for distances of 4 or 5 km is
handled by the use of inverse distance weighting in the proximity score formula.
By contrast, a buffer report, again, averages together all residents in the buffer, treating them all equally
regardless of their distance from the buffer center. Therefore, many EJSCREEN users may wish to define
a modest buffer distance that focuses on those residents who may be "most affected" and "similarly
affected" by a single facility or site of interest at the center of the circular buffer.
A very large buffer (e.g., over three miles in radius) could provide misleading results if the goal is to
describe the "affected" population, because people in a large buffer could be extremely varied in the
extent of their exposure to some source at the center of such a large buffer. If impacts decline with
distance, a large buffer would mix many relatively unimpacted, distant residents, with fewer residents
who are closer and impacted more, giving a diluted result that fails to describe those most impacted.
31 | P a g e
-------
Overview of Data and Methods
At the other extreme, a very small buffer (e.g., the size of a local block group or smaller) is problematic
because it could fail to include some significantly affected residents, and also because estimates are
more uncertain for smaller geographic areas due to sampling error in the ACS and spatial error in
estimating which residents are inside the buffer. Some EJSCREEN users may wish to define a large buffer
distance when they know a local facility or site covers a wide area (for example some NPL sites can be
very large). Some users may wish to run and compare separate buffer reports for two or more choices of
distance, or define hand-drawn buffers to look at zones at various distances from some point.
32 | P a g e
-------
Details on Environmental Indicators
3 DETAILS ON THE ENVIRONMENTAL INDICATORS IN
EJSCREEN
Environmental Factors Not Included
As described above, EJSCREEN contains 12 environmental factors, which were selected after a review of
available data, other EJ tools and analyses, and the data selection criteria discussed above.
A number of possible factors were identified in the review, which were not ultimately included in
EJSCREEN due to various limitations. These limitations were almost always a lack of high resolution data
(e.g., only available at county level), and/or lack of geographic coverage (e.g., only available in selected
or sampled locations). In some cases a factor was not added because of a high degree of overlap and
double-counting with existing indicators, or resource constraints and practical considerations. One or
more of these factors may be included in future versions of EJSCREEN as more data become available.
Other EPA resources also have more information on many of these issues, such as
EPA's website (www.epa.gov)
Envirofacts (http://www.epa.gov/enviro/)
A county-level US Environmental Quality Index (EQI)19
C-FERST (http://www.epa.gov/heasd/c-ferst/)
EnviroAtlas (http://enviroatlas.epa.gov/enviroatlas/atlas.html)
EJSCREEN also provides the ability to view some of these issues directly within the EJSCREEN maps, such
as impaired water bodies, criteria air pollutant nonattainment areas, TRI facilities, and others.
Furthermore, users can import and view, within EJSCREEN, other maps available on the internet, for
more of these environmental issues.
Users of EJSCREEN are also encouraged to consider these issues, where appropriate, to the extent they
have relevant local information.
Factors not currently used in EJSCREEN include:
Health data (e.g., overall mortality rate) - Note this is not an environmental factor, but is
sometimes of interest in this context. Relevant data were found to be available only at
county level resolution.20
Drinking water (from private wells or public water supplies) and surface water quality (other
than through the potential relevance of the NPDES proximity indicator in EJSCREEN)21
19 Messer, Jagai, Rappazzo, & Lobdell (2014)
20 Useful resources include http://www.countvhealthrankings.org, http://www.americashealthrankings.org/,
http://www.healthindicators.gov/, and http://www.rwif.org/en/research-publications/research-features/rwif-
datahub/national.html#q/scope/national/ind/31/dist/29/char/119/time/14/viz/map/cmp/brkdwn
21 See http://www2.epa.gov/learn-issues/water-resources and
http://water.epa.gov/scitech/datait/tools/waters/tools/index.cfm
33 | P a g e
-------
Details on Environmental Indicators
Contaminated fish/ seafood (other than through the potential relevance of the NPDES
proximity indicator in EJSCREEN)22
Beach closures due to pathogens (other than through the potential relevance of the NPDES
proximity indicator in EJSCREEN)23
Impaired surface waters (assessed only in certain locations, so lacking complete coverage of
all locations in the US)
Sea-level rise or other impacts of global climate change24
Radon gas exposure25 or indoor air pollutants other than radon26
Criteria air pollutants other than PM2.5 and ozone (Pb, CO, SOx and NOx)27 (Not included due
to resource constraints, and more limited modeling/coverage, frequency of nonattainment,
or health impact than for PM2.5 and ozone).
Proximity to other point or area sources not already accounted for by EJSCREEN's proximity
indicators, NATA indicators, or other air quality indicators. This can include some TRI
reporting facilities that do not emit HAPs to air, for example. TRI facilities are a small but
important fraction of all regulated facilities. Those emitting HAPs to air are already
considered through the NATA indicators, and many others are included in the RMP
indicator.28
Exposures to short episodes of elevated releases of air pollutants during startup, shutdown,
malfunction, etc. (data gaps in coverage and resolution)
Exposures to undocumented emissions caused by leaks
Exposures related to oil and gas extraction, such as hydraulic fracking29
Mining (e.g., uranium mining, etc.)30
Coal ash ponds
Combined animal feeding operations (CAFOs)31
Leaking underground storage tanks or other contaminated sites other than National
Priorities List (NPL) sites, Risk Management Plan (RMP) facilities, and Treatment, Storage
and Disposal Facilities (TSDFs)32
Pesticide exposures from spray drift or other sources, or pesticide exposures from
residential and other non-agricultural uses33
See http
//www2.epa.gov/learn-issues/water-resources
See http
//www2.epa.gov/learn-issues/water-resources
See http
//www.epa.gov/climatechange/
See http
//www.epa.gov/iaa/ia-intro. html
See http
//www.epa.gov/iaa/ia-intro. html
See http
//www.epa.gov/air/urbanair/
28 For examples of other efforts to consider a wide range of facilities, see the various TRI mapping tools such as
(http://www2.epa.gov/toxics-release-inventorv-tri-program/tri-data-and-tools). and see EPA's RSEI tool:
http://www.epa.gov/opptintr/rsei/index.html
29 See http://www2.epa.gov/hvdraulicfracturing
30 See http://water.epa.gov/polwaste/npdes/Mining.cfm
31 See http://water.epa.gov/polwaste/npdes/afo/index.cfm
32 See http://www.epa.gov/swerustl/overview.htm
33 See http://www2.epa.gov/safepestcontrol and http://www2.epa.gov/science-and-technology/pesticides-science
34 | P a g e
-------
Details on Environmental Indicators
Noise pollution and odors not already accounted for in other data34
Occupational exposures
Exposures related to imported or domestic consumer products, foods and beverages, or
other sources of exposure where we lack detailed geographic data
Ecosystem services35
EJSCREEN is designed to be a nationally consistent screening tool, with results calculated and displayed
at the Census block group level. Data inputs must be from publicly available sources, available and
consistent across the entire country, and with sufficient spatial resolution. There must be some plausible
means of quantifying an adverse effect or a proxy for an adverse effect on residential populations. These
requirements set a high bar for including environmental data. Currently, none of the potential
environmental factors listed above meet those criteria.
Environmental Factors in EJSCREEN
Each of the 12 environmental indicators included in EJSCREEN can be viewed separately. Each
environmental indicator is also combined with demographic indexes to form the EJ indexes outlined
above. Table 2 summarizes the environmental indicators in EJSCREEN, and the following sections
describe each environmental indicator in more detail. The sections above address criteria for selecting
which environmental factors to include in this version of EJSCREEN. Appendix D provides summary
statistics for each indicator, including mean and percentile values.
Table 2. Summary Table of Environmental Indicators and Sources
Key
Indicator
Details
Source [
Medium
Air
Air
Air
NATA air
toxics
cancer risk
NATA
neurological
hazard
index
NATA
respiratory
hazard
index
Lifetime cancer risk from
inhalation of air toxics
Air toxics neurological hazard
index (ratio of exposure
concentration to health-based
reference concentration)
Air toxics respiratory hazard
index (ratio of exposure
concentration to health-based
reference concentration)
EPA NATA, retrieved 20XX
http://www.epa.gov/ttn/atw/nat
amain/index.html
EPA NATA, retrieved 20XX
httpV/www.epa.p""74"4""^4"'"7^
amain/index.html
9
20xx
EPA NATA, retrieved 20XX
http://www.epa.gov/ttn/at
amain/index.html
Air
NATA diesel
PM
Diesel particulate matter level
in air, ng/m3
EPA NATA, retrieved 20XX
http://www.epa.gov/ttn/atw/nat
amain/index.html
20xx
34 See http://www.epa.gov/air/noise.html
35 See http://enviroatlas.epa.gov/enviroatlas/
35 | P a g e
-------
Details on Environmental Indicators
Air
Particulate
matter
PM2.5 levels in air, ng/m3
annual avg. (2011)
EPA, OAR (fusion of model and
monitor data). For methods, see
http://www.epa.gov/esd/land-
sci/lcb/lcb faasd.html
2011
Air
Ozone
Ozone summer seasonal avg.
of daily maximum 8-hour
concentration in air in parts
per billion (2011)
EPA, OAR (fusion of model and
monitor data). For methods, see
http://www.epa.gov/esd/land-
sci/lcb/lcb faqsd.html
2011
Air/other
Traffic
proximity
and volume
Count of vehicles (AADT, avg.
annual daily traffic) at major
roads within 500 meters,
divided by distance in meters
(not km)
Calculated from 2011 U.S. DOT
traffic data, retrieved 4/2012
http://www.rita.dot.gov/bts/site
s/rita.dot.gov. bts/files/publicatio
ns/national transportation atlas
database/2011/index.html
2011
Dust/
lead
paint
Lead paint
indicator
Percent of housing units built
pre-1960, as indicator of
potential lead paint exposure
Calculated based on Census/ACS
data, retrieved 2014
http://www2.census.gov/acs201
2008-
2012
2 5vr/summarvfile/
Waste/
air /
water
Proximity to
RMP sites
Count of RMP (potential
chemical accident
management plan) facilities
within 5 km
(or nearest one beyond 5 km),
each divided by distance in
kilometers
Calculated from EPA RMP
database, retrieved 11/2013
2013
Waste/
air /
water
Proximity to
TSDFs
Count of TSDFs (hazardous
waste management facilities)
within 5 km (or nearest
beyond 5 km), each divided by
distance in kilometers
Calculated from EPA RCRAInfo
database, retrieved 11/2013
http://www.epa.gov/enviro/fact
s/rcrainfo/search.html
2013
Waste/
air /
water
Proximity to
NPL sites
Count of proposed and listed
NPL sites36 within 5 km (or
nearest one beyond 5 km),
each divided by distance in
kilometers
Calculated from EPA CERCLIS
database, retrieved 11/12/2013
http://cumulis.epa.gov/supercpa
d/cursites/srchsites.cfm
2013
Water
Proximity to
major direct
water
dischargers
Count of NPDES major direct
water discharger facilities
within 5 km (or nearest one
beyond 5 km), each divided by
distance in kilometers
Calculated from EPA PCS/ICIS
database, retrieved 12/2013
http://www.epa.gov/enviro/fact
s/pcs-icis/search.html
2013
Note: EJSCREEN's EJ Indexes also include demographic information that is obtained from the U.S.
Census Bureau's American Community Survey (ACS). The 2015 version of EJSCREEN includes 2008-
2012 ACS 5-year summary file data, which is based on 2010 Census boundaries.
36 Count of NPL sites excludes deleted sites, sites in U.S. territories, and other sites that could not be included.
36 | P a g e
-------
Details on Environmental Indicators
NATA Air Toxics and NATA Diesel PM
Air toxics, often referred to as hazardous air pollutants (HAPs), are pollutants that are known or
suspected to cause cancer or other serious health effects, such as reproductive effects or birth defects,
or adverse environmental effects. EPA regulates 187 chemicals under its HAP program (U.S. EPA, 2009d).
Most air toxics originate from transportation and industry, including motor vehicles, industrial facilities
and power plants.
Indicator EPA's National Air Toxics Assessment (NATA) provides the following indicators
that are used in EJSCREEN:
Estimated lifetime inhalation cancer risk from the analyzed carcinogens in
ambient outdoor air.
Hazard index for respiratory effects.
Hazard index for neurological effects.
Diesel particulate matter concentration.
Rationale for A chemical's listing as a HAP is based on evidence of cancer or other adverse
Inclusion health effects or environmental effects associated with exposure to the chemical,
as determined by EPA and the initial list in the Clean Air Act Amendments of
1990. EPA's Integrated Risk Information System (IRIS) program documents the
health risks associated with these chemicals and serves as a basis for the analysis
of health implications (U.S. EPA, 2012c).
Air toxics cancer risk and noncancer impacts have been included in other EPA EJ
screening tools.
HAPs are emitted from a wide variety of sources and disperse around the sources,
especially downwind. In some cases, these substances react with other
constituents in the atmosphere or break down to other chemicals, and most are
eventually removed through precipitation or other atmospheric processes. People
are exposed in their daily activities in and around their homes, at school or work,
and while moving about the area. They inhale the substances, exhale or excrete
some portion of them, and have the potential for incurring adverse effects from
the portion that stays in the body.
More Information More information is available at the air toxics website
(http://www.epa.gov/air/toxicair), the NATA website (www.epa.gov/nata), and
the IRIS website (www.epa.gov/iris).
Relevant Studies A comprehensive list of EJ studies using the NATA database can be found in
Chakraborty et al., (2011). Some examples of EJ studies of chemicals listed as
HAPs include Morello-Frosch & Jesdale (2006), and other studies reviewed by Liu
37 | P a g e
-------
Details on Environmental Indicators
Data Source
Data Version
(2001) and Brender et al., (2011). Diesel particulate matter has also been the
subject of EJ analysis (Rosenbaum, Hartley, & Holder, 2011).
EJSCREEN uses the most recent data from EPA's National-Scale Air Toxics
Assessment (NATA). NATA estimates cancer risk or noncancer implications of
many of the 187 air pollutants classified as HAPs, as well as diesel particulate
matter. NATA uses emissions estimates from the National Emissions Inventory
(NEI), which is updated every three years. The NEI includes all of the Toxics
Release Inventory (TRI) reporting facilities that release hazardous air pollutants,
along with many other sources of air pollutants, such as motor vehicles.
Note that the publicly-available NATA, PM2.5, and ozone estimates are at tract
resolution, and tract level is the resolution used for EJSCREEN, unlike with
proximity indicators, for example. Each block group was assigned the NATA or PM
or ozone score of the tract containing it. All indicators or statistics then were
calculated using block group data, whether or not those block group scores had
been assigned based on tracts.
The 2015 version of EJSCREEN uses 20xx NATA data, which is based on NEI
emissions estimates for 20xx (U.S. EPA, 2015b). This version of NATA estimated
ambient concentrations of 177 HAPs plus DPM, and then estimated health
implications for 139 of these HAPs (cancer risk for 80 and noncancer results for
100).
NATA update pending
Data from recent years may no longer be as representative of current conditions
as they were at the time the data was collected. The NATA-based indicators in
particular should be viewed with this in mind, because emissions of air toxics
generally have decreased in the intervening years. This version of EJSCREEN
incorporated the most recent data that were available at the time of indicator
development. Every attempt will be made to use the most recent appropriate
data available in future updates of EJSCREEN. There is always a delay between the
release of raw data and their eventual incorporation into any models, tools, or
maps. It is also useful to note that although the raw numbers for some indicators
do not represent current conditions, the percentiles are much more likely to be
reasonably representative of today's conditions in most locations. This is because
even if emissions have been significantly reduced overall, for example, the
differences between various locations are unlikely to have changed as
dramatically, especially when the reductions have come from national regulations
and other trends affecting entire industries or sectors in many locations. For this
reason, the percentiles may be more representative of current conditions than
the raw values of the indicators.
38 | P a g e
-------
Details on Environmental Indicators
Discussion EPA's NATA website has extensive documentation of all of the data and methods
used in developing the NATA indicators, as well as discussions of uncertainty,
caveats, and limitations in the NATA estimates. That information is not repeated
here, but it is important that anyone using NATA data understand these issues, so
anyone using EJSCREEN should consult the NATA documentation
(www.epa.gov/nata).
Very briefly, the air pollutants in NATA include likely or known carcinogens such
as formaldehyde, benzene, polycyclic aromatic hydrocarbons and naphthalene, as
well as important sources of noncancer impact such as acrolein, DPM and
manganese compounds. The cancer risk in NATA is aggregated as a cumulative
risk for the combination of all analyzed HAPs, and this total is used in EJSCREEN.
NATA calculates a hazard quotient, which is the ratio of ambient air concentration
to a chemical's health-based RfC. No adverse health effects are expected from
exposure if the hazard quotient is less than one. A hazard index is the sum of
hazard quotients for chemicals that cause adverse effects through the same toxic
mechanism. NATA currently includes hazard indexes for respiratory effects and
neurological effects, both of which are included as environmental indicators in
EJSCREEN. Each represents the cumulative impacts of all the relevant air toxics.
The NATA website provides more detailed data than EJSCREEN - Tables and maps
on individual HAPs or specific types of sources (e.g., mobile sources only) can be
generated by GIS practitioners using data from the NATA website, for those
requiring more detail than is provided by the data in EJSCREEN.
The reports in EJSCREEN present the environmental indicators from NATA using
ranges of percentiles such as 90-95 or 95-100 rather than as the numbers 1-100
(for Regional and US percentiles). This is done in recognition of the uncertainties
inherent in comparing NATA estimates across States that may have different
approaches in emissions inventories.
39 | P a g e
-------
Details on Environmental Indicators
Particulate Matter (PM2.5)
PM2.5 is particulate matter that is 2.5 microns or less in diameter. Common sources of PM2.5 emissions
include power plants and industrial facilities. Secondary PM2.5 can form from gases, such as oxides of
nitrogen (NOx) or sulfur dioxide (S02), reacting in the atmosphere. EPA set the first PM2.5 National
Ambient Air Quality Standards (NAAQS) in 1997, and revised the standards in 2006 and 2012.
Indicator Annual average PM2.5 concentration in micrograms per cubic meter
(Hg/m3).
Rationale for EPA's work associated with the PM NAAQS has documented the health
Inclusion effects associated with exposure to PM2.5, including elevated risk of
premature mortality from cardiovascular diseases or lung cancer, and
increased health problems such as asthma attacks (U.S. EPA, 2009b).
PM2.5 concentrations at different levels are found in all parts of the United
States, so residents are exposed via inhalation to varying degrees. A 2012
EPA report found that the majority of the U.S. population lived in areas in
nonattainment of one or more of the NAAQS in effect at that timea total
population of 159 million people (based on 2010 Census data for those
locations) (U.S. EPA, 2012g).
Several studies relevant to EJ and PM2.5, including those discussing
susceptible subgroups, are referenced by Bell & Ebisu (2012).
PM2.5 has been included in other EPA EJ screening tools.
More Information More information is available at the PM2.5 website
(http://www.epa.gov/pm).
Relevant Studies Some examples of studies focused on disparities in exposure to PM2.5 have
been reviewed in Liu (2001), and more recent studies include Bell & Ebisu
(2012); Fann et al. (2011); Post, Belova, & Huang (2011); Miranda,
Edwards, Keating, & Paul (2011); Brochu et al. (2011); and Levy, Wilson, &
Zwack (2007). A very recent study found disparities in exposure to NOx, a
precursor to PM2.5- (Clark, Millet, & Marshall, 2014).
Data Source EJSCREEN's PM2.5 data are estimated from a combination of monitoring
data and air quality modeling. Ambient PM2.5 concentration is estimated by
EPA's Office of Research and Development using a Bayesian space-time
downscaling fusion model approach. This approach is described in a series
40 | P a g e
-------
Details on Environmental Indicators
of three published journal articles (Berrocal, Gelfand, & Holland, 2010a,
2010b, 2011).37
PM2.5 and ozone estimates were not available for Alaska or Hawaii for use
in the 2015 version of EJSCREEN, due to a lack of CMAQ modeling. EPA
may be able to include estimates in a future version of EJSCREEN.
Data Version The 2015 version of EJSCREEN uses PM2.5 data that are based on 2011
monitoring and modeling estimates (U.S. EPA, 2015a).
Data from several years ago may no longer be as representative of current
conditions as they were at the time the data was collected. The PM2.5 and
ozone indicators in particular should be viewed with this in mind, because
emissions related to PM2.5 and ozone generally have decreased in the
intervening years. This version of EJSCREEN incorporated the most recent
data that were available at the time of indicator development. Every
attempt will be made to use the most recent appropriate data available in
future updates of EJSCREEN. There is always a delay between the release
of raw data and their eventual incorporation into any models, tools, or
maps. It is also useful to note that although the raw numbers for some
indicators do not represent current conditions, the percentiles are much
more likely to be reasonably representative of today's conditions in most
locations. This is because even if emissions have been significantly reduced
overall, for example, the differences between various locations are unlikely
to have changed as dramatically, especially when the reductions have
come from national regulations and other trends affecting entire industries
or sectors in many locations. For this reason, the percentiles may be more
representative of current conditions than the raw values of the indicators.
Finally, some supplementary maps and local information can complement
the EJSCREEN indicators to provide more recent information. In particular,
EJSCREEN also provides updated maps of PM2.5 and ozone nonattainment
areas (areas not meeting national ambient air quality standards).
Resolution High-resolution estimates of PM2.5 are very difficult to develop for the
entire United States. Block groups vary widely in geographic areasome
are larger than 100 square kilometers, but a substantial fraction are
smaller than 1 sq. km in area. This makes it challenging to develop relevant
spatial data.
37 Detailed documentation and GIS metadata describing a tract-level application of the same approach are
provided here: http://www.epa.gov/esd/land-sci/lcb/lcb faqsd.html and http://www.epa.gov/esd/land-
sci/lcb/pdf/DSMetadataAir 0612.pdf.
41 | P a g e
-------
Details on Environmental Indicators
Some small areas have used the high-resolution AERMOD model to
estimate PM2.5 levels for EJ analysis (Maroko, 2012), but such modeling is
not feasible for the entire United States at this time.
In the past (prior to approximately 2007), CMAQ used a grid size of 36x36
km in the Western United States and 12x12 km in the Eastern United
States, or 36 km in general, as in a recent EJ analysis of the heavy-duty
diesel emissions rule finalized in 2001 (Post et al., 2011). After
approximately 2007, CMAQ modeling has divided the nation into a grid of
cells that are each roughly 12 km by 12 km, and has estimated the PM2.5
concentration in each cell.
In a 2010 study, satellite data have been used to estimate PM2.5 levels with
a spatial resolution of roughly 10 km by 10 km, showing reasonably good
agreement with monitoring data (van Donkelaar et al., 2010).
Land use regression (LUR) has also been proposed, and can provide better
resolution. Nationwide LUR-based estimates have been developed for NOx
but not PM2.5.
The downscaler method was selected for EJSCREEN partly because it is
particularly useful for this application, in that it estimates concentration at
a specified point, rather than for the average of a large grid cell. The
downscaler algorithms combine information from nearby monitors and
CMAQ grid cell estimates. This provides an estimate based on more
information than models alone or monitors alone could provide. It is
important to note that the downscaler and indicators here are not
attempting to describe all of the local variations in ambient air
concentrations. They are merely capturing some additional variation that is
not seen when relying on models or monitors alone.
Discussion The downscaling fusion model uses both air quality monitoring data from
NAMS/SLAMS (data collected by EPA, state, local and tribal air pollution
control agencies at more than 600 hundred monitors nationwide) and
numerical output from the Models-3/CMAQ model. The CMAQ model is
used extensively by EPA and has been described in detail elsewhere (Byun
& Schere, 2006).
This downscaling approach is designed to provide daily, predictive PM2.5
(daily average) and 03 (daily 8-hour maximum) surfaces for a given year,
such as 2011, at specified points.
For EJSCREEN, the downscaling method was applied to a point within each
Census tract. EPA's Office of Air and Radiation generated an estimate for
42 | P a g e
-------
Details on Environmental Indicators
each tract, and then assigned the same tract value to every block group
within the given tract. Daily estimates from the downscaling method were
averaged for the whole year in the case of PM2.5 and for the ozone season
(May-September) in the case of ozone.
Again, it is important to note that the downscaler and indicators here are
not attempting to describe all of the local variations in ambient air
concentrations. They are merely capturing some additional variation that is
not seen when relying on models or monitors alone.
Several data sources have been used elsewhere and were considered for
inclusion in EJSCREEN. For instance, EPA's regulatory impact analyses
(RIAs) for recent rules and the PM2.5 NAAQS have used estimates of PM2.5
that combine modeling and monitoring in a different way (U.S. EPA,
2009b). These estimates also start with CMAQ air quality modeling results,
but then locally adjust those estimates up or down based on local
monitoring data using MATS (monitor attainment test software), which
provides an enhanced Voronoi neighbor averaging interpolation technique.
Published analyses of PM2.5 health impacts have used similarly fused
estimates (Fann et al., 2012). Before the downscaler method was
developed, a different Bayesian modeling approach was also used, as
described in McMillan, Holland, Morara, & Fang (2010). Other efforts have
used interpolation between monitors without air quality modeling, such as
through basic Voronoi neighbor averaging (Fann & Risley, 2011), or simply
the average of monitors within a county (Bravo, Fuentes, Zhang, Burr, &
Bell, 2012). Monitors provide reliable estimates where they are located,
but suitable PM2.5 data are available at fewer than 900 monitors in the
United States. While urban areas tend to have PM2.5 monitors, more than
two-thirds of U.S. counties lack any monitoring data, so modeling is an
important complement to monitoring. Methods based on CMAQ alone,
monitors alone, CMAQ-MATS and the downscaling approach all provide
somewhat different estimates.
Note that the EJSCREEN value does not directly indicate nonattainment of
the NAAQS standard because the indicator in EJSCREEN is based on
estimates from a combination of modeling and monitoring for a single
year, while nonattainment is determined for a large area (often a county)
based on three years of monitoring data.
43 | P a g e
-------
Details on Environmental Indicators
Ozone
Ozone (03) is not usually emitted directly into the air, but is created at ground level by a chemical
reaction between oxides of nitrogen (NOx) and volatile organic compounds (VOCs) in the presence of
sunlight. These ozone precursors are emitted by motor vehicles, industrial facilities and power plants as
well as natural sources. Ground-level ozone is the primary constituent of smog.
Indicator The May-September (summer/ ozone season) average of daily-maximum
8-hour-average ozone concentrations, in parts per billion (ppb).
Rationale for Toxicological and epidemiological studies have established an association
Inclusion between exposure to ambient ozone and a variety of health outcomes,
including reduction in lung function, increased inflammation and increased
hospital admissions and mortality (U.S. EPA, 2006b). In the 2006 Air Quality
Criteria Document for Ozone, a comprehensive review of the clinical and
epidemiological evidence was inconclusive about a possible threshold for
ozone-induced health effects. EPA concluded that if a population threshold
level exists, it is near the lower limit of ambient ozone concentrations in
the United States (U.S. EPA, 2006b). Several subpopulations may
experience susceptibility to ozone-induced health effects. These
subpopulations include older adults, children, individuals with preexisting
pulmonary disease and those with higher exposure levels such as outdoor
workers (U.S. EPA, 2006b). A recent review of studies identifying
subgroups susceptible to ozone found the strongest evidence for greater
sensitivity among the elderly and also the unemployed (Bell, Zanobetti, &
Dominici, 2014).
A 2012 EPA report found that the majority of the U.S. population lived in
areas in nonattainment of one or more of the NAAQSa total population
of 159 million people (based on 2010 Census data for those locations) (U.S.
EPA, 2012g). As standards are updated, nonattainment areas are redefined
along with the number of people living in redefined nonattainment areas.
Ozone concentrations at different levels are found in all parts of the United
States, so residents are exposed via inhalation to varying degrees.
Ozone has been included in other EPA EJ screening tools.
More Information More information is available at the ground-level ozone website
(http://www.epa.gov/air/ozonepollution/).
Relevant Studies Some examples of studies that have focused on disparities in exposure to
ozone include Fann et al. (2011), Grineski (2007), Grineski et al. (2007), and
those reviewed in Liu (2001).
44 | P a g e
-------
Details on Environmental Indicators
Data Source
Data Version
Resolution
Discussion
45 | P a g e
EJSCREEN's ozone data are estimated by EPA from a combination of
monitoring data and CMAQ air quality modeling. Ozone was estimated
with the same approach as PM2.5, and the methodology is described above.
Ozone faces similar limitations, in that a limited number of U.S. monitors
have suitable data, so modeling is an important complement to monitoring
data.
PM2.5 and ozone estimates were not available for Alaska or Hawaii for use
in the 2015 version of EJSCREEN, due to a lack of CMAQ modeling. EPA
may be able to include estimates in a future version of EJSCREEN.
The 2015 version of EJSCREEN uses ozone data that are based on 2011
monitoring and modeling estimates (U.S. EPA, 2015a). Tract estimates
were assigned to block groups as described for the PM2.5 indicator.
See "Resolution" section under the previous environmental indicator,
PM2.5.
See "Discussion" section under the previous environmental indicator,
PM2.5.
-------
Details on Environmental Indicators
Lead Paint Indicator
A key source of exposure to lead for many Americans is through lead paint and lead-containing dust that
accumulates indoors, in homes or in other buildings where lead paint was used. Exterior structures
painted with lead-based paint are also a source of ambient lead through chipping exterior paint.
Elevated short-term lead dust loadings have also been observed following demolition of old buildings
(U.S. EPA, 2011c). Lead-based paint was banned in the United States by the Consumer Product Safety
Commission in 1978, but lead-based paint used in housing before the ban remains a significant source of
exposure to lead for children and adults. Lead paint and contaminated dust and soil are considered the
leading cause of high lead levels in U.S. children (Levin et al., 2008).
Indicator The percentage of occupied housing units built before 1960 was selected
as an indicator of the likelihood of having significant lead-based paint
hazards in the home.
Rationale for Elevated blood lead levels are a well-documented public health concern
Inclusion of particular interest to EJ stakeholders, and represent an important
environmental health issue (U.S. EPA, 2006a, 2011c).
Certain demographic groups may be more susceptible to lead exposure.
For example, blood lead's association with cardiovascular outcomes
appears to be stronger among Mexican Americans and non-Hispanic
blacks than non-Hispanic whites (U.S. EPA 2011c). Also, some but not all
studies suggest lead has a greater impact on IQ among those of low
socioeconomic status (U.S. EPA 2011c).
Despite significant reductions in ambient levels of lead from the phase-
out of leaded gasoline and the 1978 ban on lead-based paint, lead
remains a significant hazard for children. Recent research has
demonstrated that children can experience neurological damage even at
low levels of exposure to lead. In May 2012, the Centers for Disease
Control and Prevention (CDC) agreed to adopt the recommendations of
the CDC Advisory Committee for Childhood Lead Poisoning Prevention
(ACCLPP) for defining elevated blood-lead levels (BLLs) among children.
The ACCLPP recommended that CDC use a childhood BLL reference value
based on the 97.5th percentile of the population BLL in children under age
6 to identify children and environments associated with lead-exposure
hazards (Centers for Disease Control and Prevention, 2012). The 97.5th
percentile value is currently 5 ng/dL.
Surfaces originally covered with lead-based paint may chip, flake or
develop a chalky surface. The lead in these pieces or particles may be
moved about the interior or exterior of the painted structures, and be
46 | P a g e
-------
Details on Environmental Indicators
moved from inside to outside and vice versa. Through direct contact with
the painted surfaces or through contact with the released particles, lead
may adhere to hands and other parts of the residents' bodies, and people
may ingest some portion of the lead. If the painted surfaces are disturbed
through renovation or other actions, some lead-based paint particles may
be temporarily suspended in the air, and particles on surfaces within the
structures may be re-suspended during the residents' activities. The
suspended particles may be inhaled or may fall on food and be ingested.
Children playing inside or outside and exposed to particles of lead-based
paint may ingest some of the lead through hand-mouth actions.
An analysis of data collected during the 1999-2004 National Health and
Nutrition Examination Survey (NHANES) showed that children living in
older housing stock (built before 1950) were significantly more likely to
have blood lead levels greater than 5 ng/dL than children living in housing
built after 1978 (Jones et al., 2009). Jones et al. estimated that 7.4% of
children under age 6 had blood lead levels greater than 5 ng/dL during
NHANES 1999-2004. For children under age 6 living in the highest risk
housing (built before 1950), Jones et al. observed that 15.1% had blood
lead levels above 5 ng/dL. For children under age 6 living in the lowest risk
housing (built in 1978 or later), 2.1% had BLLs above 5 ng/dL.
EPA EJ screening tools in the past generally have not included proxies for
lead exposure.
More Information More information is available at EPA's lead website
(http://www2.epa.gov/lead).
Relevant Studies Several examples of EJ studies of exposure from lead paint exist, including
Gaitens et al., 2009 and others.
Data Source The data were collected at the block group level from the ACS estimates
from the Census Bureau. The indicator was calculated by dividing the
count of occupied housing units built prior to 1960 by the total number of
built housing units in the block group (ACS table B25034, see Appendix B
for details).
Data Version The block-group level data for the 2015 version of EJSCREEN were
collected from the 2008-2012 ACS (U.S. Census Bureau, 2011).
Approximately 40% of occupied, non-institutional housing units in the
United States were built prior to 1960, as of 1999. The ACS 2008-2012
47 | P a g e
-------
Details on Environmental Indicators
data in EJSCREEN indicate the average person in the US lives in a block
group where about 30% of the housing was built before 1960.
Discussion Lead paint was used extensively in the United States prior to the 1978 ban
on lead in new residential paint, and a home built prior to 1960 is far
more likely to have lead hazards than one built more recently (Gaitens et
al., 2009; Jacobs et al., 2002). In 2002, Jacobs et al. reported that
approximately 40 million homes in the United States still had lead-based
paint hazards, based on a nationally representative survey conducted in
1998-2000. The likelihood of such hazards was found to have changed
dramatically for housing built in 1960-1977 compared to pre-1960
housing (Table 3).
Based on Jacobs et al. (2002), EPA calculated the following likelihoods of
significant lead-based paint hazards:
Pre-1960 vs. all others: 54% vs. 6% (9 times as likely).
Pre-1960 vs. all others, among those with children under age 6:
68% vs. 4% (16 times as likely).
Data and analysis published in 2009 confirmed prior conclusions that
potential exposure to lead is associated with housing age, providing more
information on lead concentrations in household dust as a function of
housing age (Gaitens et al., 2009). Some of the models presented by
Gaitens et al. (2009) suggest that the largest decreases in lead dust levels
are seen between housing built prior to 1940 and after 1940, with more
modest contrasts seen for housing built after 1960 and after 1977. A
cutoff of 1960, however, is consistent with the data from Jacobs et al.
(2002), and the window sill lead dust models from Gaitens et al. (2009).
It is important to note that older housing alone may not represent any
actual risk or even exposure.
48 | P a g e
-------
Details on Environmental Indicators
Table 3. Likelihood of Lead-Based Paint Hazards by Housing Construction Date
Share of Housing with Significant Lead-Based
Year Built Paint Hazards (and 95% Confidence Interval)
Post-1960:
1978-1998
3% (1-6%)
1960-1977
8% (6-12%)
Pre-1960:
1940-1959
43% (32-51%)
Before 1940
68% (56-75%)
Source: Jacobs et al. (2002). A "significant lead-based paint hazard" is defined as "a lead-based paint hazard above de
minimis levels as defined in U.S. EPA and U.S. Department of Housing and Urban Development (HUD) regulations."38
38 The de minimis levels for paint deterioration are < 20 ft2 (exterior) or < 2 ft2 (interior) of lead-based paint on
large surface area components (walls, doors), or damage to < 10% of the total surface area of interior small surface
area component types (windowsills, baseboards, trim) (40 C.F.R. § 745.65).
49 | P a g e
-------
Details on Environmental Indicators
Traffic Proximity
A substantial fraction of the U.S. population lives in close proximity to traffic, and the number of vehicle-
miles traveled has grown 40% from 1990 to 2010 (U.S. EPA, 2012d). Proximity to motor vehicle traffic is
associated with increased exposures to ambient noise, toxic gases and particulate matter including
diesel particulates. Technical details about the methodology used to determine traffic proximity are
provided in Appendix C.
Indicator The count of vehicles per day within 500 meters of a block centroid, divided by
distance in meters, presented as the population-weighted average of blocks in each
block group. Adjustments are made so that the minimum distance used is
reasonable when very small.
Rationale for A 2011 literature review identified several studies that "found that living near
Inclusion hazardous waste sites, industrial sites, cropland with pesticide applications, highly
trafficked roads, nuclear plants, and gas stations or repair shops is related to an
increased risk of adverse health outcomes" (Brender et al., 2011, p. S37). This
indicator is the first of five that relate to this category of concern.
Is should first be noted that there are both positive and negative aspects to living
near major roads. Proximity to roads can provide access to jobs, health care, food,
recreational opportunities, and other benefits. The indicator of traffic proximity and
volume is designed to screen for the negative aspects, so it uses distance weighting
and volume weighting to focus on the extremes of very close proximity to very high
volumes of traffic, such as living closer than 50-100 meters from a multi-lane
highway, as explained below.
Residential proximity to traffic has been associated with various health impacts,
particularly asthma exacerbation and possibly onset of asthma, as well as mortality
rates (Baumann et al., 2011; Health Effects Institute, 2010). Proximity to traffic has
also been associated with subclinical atherosclerosis (a key pathology underlying
cardiovascular disease (CVD)), prevalence of CVD and coronary heart disease (CHD),
incidence of myocardial infarction, and CVD mortality (Hoffman et al., 2009).
Vehicle-related emissions of various pollutantsultrafine and other components of
PM2.5, lead and other metals, and mobile source air toxics such as benzene, nitrogen
oxides (NOx), hydrocarbons and carbon monoxide (CO)are believed to contribute
to these health effects. Vehicles also emit precursors that add to ambient ozone
and PM2.5. Additionally, EPA's 2005 NATA estimated that mobile emissions
accounted for about 30% of average cancer risk from the pollutants in NATA, mainly
from benzene (U.S. EPA, 2009c). However, the spatial accuracy of NATA's mobile
source impacts is limited, because local estimates are based on countywide total
mobile source emissions roughly allocated to each part of the county based on
50 | P a g e
-------
Details on Environmental Indicators
presence of major roads. The traffic indicator in EJSCREEN provides a more detailed
analysis of the volume and location of traffic than was used in NATA. Also, NATA
captures only some of the impacts associated with traffic, so the traffic indicator is a
useful complement.
Traffic proximity is also associated with noise, which is a risk factor for various
health problems. Workplace and transportation-related noise have been associated
with release of stress hormones; sleep disturbance; hypertension; altered heart
rate; ischemic heart disease; myocardial infarction; and, among the elderly, risk of
stroke (S0rensen et al., 2011). In one study, for example, among those older than
64.5 years of age, the stroke incidence rate ratio was 1.27 per 10 dB more road
traffic noise (S0rensen et al., 2011).
Whether noise or other factors account for it, local traffic volume is a predictor of
stress (which itself is associated with significant health risks). In 2010, Yang &
Matthews concluded that, "[a]t the neighborhood level, the presence of hazardous
waste sites and traffic volume were determinants of self-rated stress even after
controlling for other individual characteristics" (2010, p. 803).
Any indicator of residential proximity addresses exposures relevant to the
residences within a block group, and would not capture most exposures that occur
away from the home, such as at work, at school or during a commute.
In the past, EJ screening tools at EPA have not included traffic proximity.
More More information is available at the near-roadway website
Information (http://www.epa.gov/otaq/nearroadway.htm).
Relevant Some examples of studies of disparities in proximity to traffic include Tian et al.,
Studies 2013; Rowangould, 2013; Brender et al., 2011; Chakraborty (2006); and Liu, 2001.
Data Source Measures of traffic proximity in EJSCREEN are based on average annual daily traffic
(AADT) estimates in the Highway Performance Monitoring System (HPMS) dataset
in the Department of Transportation (DOT) National Transportation Atlas Database
(NTAD). The HPMS highway data is maintained by states and compiled by DOT.
The HPMS data are collected at the state level, and the traffic counting program is
designed to cover all interstate, principal arterial, other National Highway System
and HPMS sample sections on a 3-year maximum cycle where at least one-third of
roads are counted each year. More details on the HPMS are available at
http://www.fhwa.dot.gov/policvinformation/hpms.cfm.
51 | P a g e
-------
Details on Environmental Indicators
Data Version The 2015 version of EJSCREEN uses 2011 HPMS data (U.S. Department of
Transportation, 2011)39. This dataset contains 490,781 road segments, of which
86.7% contain AADT information, and those have a length of approximately
548,000 km (340,511 miles). While this is only 8% of total road miles of public road
in the US, the roads not included (e.g., local streets) tend to have much lower levels
of traffic, so the roads included appear to account for almost two thirds of all US
traffic (vehicle-miles travelled).40 For the 2015 version of EJSCREEN, a total of
11,078,297 Census 2010 blocks were analyzed to find all road segments within 500
meters of each block's internal point, or the nearest single segment if none were
found within 500 meters.
Discussion The traffic proximity indicator is based on average annual daily traffic (AADT)
divided by distance in meters. For example, a single highway with 16,000 AADT at
400 meters distance would result in a score of 16,000/400=40, which is close to the
median person's block group traffic proximity indicator value in the US. About 5% of
the population has traffic proximity indicator values more than ten times as high as
the median, because traffic volumes vary widely across roads and communities. The
most traveled highway section in the United States, the 1-405 in the Los Angeles-
Long Beach-Santa Ana area, had 374,000 vehicles of AADT in 2008.41 About forty
other highway sections in the US exceeded 250,000 AADT, but about half were in
just one state - California - and the were rest spread over just a dozen states.
The proximity score is based on the traffic within a search radius of 500 meters (or
further if none is found in that radius). It is important to understand that this
distance was selected to be large enough to capture the great majority of road
segments (with traffic data) that could have a significant impact on the local
residents, balanced against the need to limit the scope due to computational
constraints. The closest traffic is given more weight, and the distant traffic given
less weight, through inverse distance weighting. For example, traffic 500 meter
away is given only one tenth as much weight as traffic 50 meters away.
Based on the available evidence, a distance of roughly 100-300 meters, or perhaps
up to 500 meters, from traffic should be considered as most important. This
distance focuses on the types of exposures typically studied and shown to be
39
http://www.rita.dot.gov/bts/sites/rita.dot.gov.bts/files/publications/national transportation atlas database/201
1/index.html and also see http://www.fhwa.dot.gov/policvinformation/hpms.cfm
40 The included 340,000 miles is comparable to the road miles covered by all interstate, freeway/expressway, and
principal arterials, plus 50% of the minor arterial miles in the US, which together carry 64% of VMT. Collector and
local roads are the balance of public roads. See
http://www.fhwa.dot.gov/policvinformation/statistics/2011/hm220.cfm and
http://www.fhwa.dot.gov/policvinformation/statistics/2011/vm202.cfm
41 Office of Highway Policy Information, US DOT, 2008 Highway performance monitoring system (HPMS) July 27,
2010.
52 | P a g e
-------
Details on Environmental Indicators
associated with health impacts in near-roadway epidemiology. Epidemiologic
studies of the impacts of proximity to traffic often utilize distances of 50-1,500
meters to define a cutoff between less and more exposed locations (Health Effects
Institute, 2010). For example, a major study of coronary heart disease prevalence
used distances greater than 200 meters as the reference group and found adjusted
odds ratios of 1.08 for residences within 100-200 meters, 1.71 for 50-100 meters
and 1.55 within 50 meters of a major road. Only 15% of participants lived within
200 meters of a major road, and only 3% within 50 meters in this study of heart
disease (Hoffman et al., 2009).
Additionally, a distance cutoff of 500 meters captures exposures of concern for
most definitions of mobile source impact. In a review of numerous prior studies of
proximity to roads, in combination with a modeling case study, Zhou & Levy (2007)
suggested that a distance of 500 meters should capture exposures of concern,
although impacts may be largely limited to just 100 meters from roads for ultrafine
particles and PM2.5 mass from mobile sources alone.
A critical review of literature on traffic-related air pollution in 2010 "identified an
exposure zone within a range of up to 300 to 500 meters from a highway or a major
road as the area most highly affected by traffic emissions... and estimated that 30%
to 45% of people living in large North American cities live within such zones"
(Health Effects Institute, 2010, p. 7-5). A 2009 analysis of PM2.5 levels in Southern
California found that traffic within 300 meters of a monitor was the most
informative predictor of monitored PM2.5 levels, out of a wide range of factors
considered such as various distances from roads, population density and the
presence of industry (Krewski et al., 2009).
On the other hand, some studies have shown a dramatic drop in at least ultrafine
levels within the first 100 meters downwind from a freeway, and an even sharper
(essentially immediate) drop in the upwind direction (Zhu, Hinds, Kim, & Sioutas,
2002). This pattern has been seen in more recent measurementslevels on
California highways (measured using monitors on vehicles) were compared to levels
near those roads (roughly 50-300 meters away), and black carbon levels in
particular were as much as 10 times higher on the road than near the road, for 1-
hour averages (Fujita, Campbell, Zielinska, Arnott, & Chow, 2011). The same study
found much higher levels (generally 2-5 times higher) on the road than near the
road, for PM2.5 mass, CO, NO, NOx, VOCs, benzene, toluene, ethylene, xylene,
formaldehyde and acetaldehyde. This reinforces the idea that exposures very close
to a busy highway are most important, and that levels drop rapidly within tens of
meters, falling to much lower levels within the first 50-300 meters (Spengler et al.,
2011).
53 | P a g e
-------
Details on Environmental Indicators
Many studies have analyzed roadway proximity categorically, including only major
roads, but roads vary in the amount of traffic they carry, so AADT provides a better
starting point for considering impacts than simply whether a road is a major road.
Several land use regression studies and other research (Hoek et al., 2011; Hystad et
al., 2011) have suggested that inverse distance-weighted traffic volume is a
reasonably good proxy for ambient air concentrations of NOx, PM2.5 mass (ng/m3),
black carbon or ultrafine PM (as particle number concentration) nearby (50-500
meters). Levels are clearly higher downwind of the road, and higher where wind
speeds are lower, but in the absence of detailed location-specific data, traffic
volume and distance are useful indicators (Hoek et al., 2011).
54 | P a g e
-------
Details on Environmental Indicators
Proximity to Major Direct Water Dischargers
The Clean Water Act regulates facilities that discharge pollutants from point sources to waters of the
United States. Discharging facilities are generally prohibited unless authorized by a specific provision of
the act. Direct discharges may be authorized by National Pollutant Discharge Elimination System
(NPDES) permits issued by EPA or states authorized to administer the NPDES program. There are tens of
thousands of dischargers with such permits, many releasing only limited quantities of pollutants. Several
thousand facilities, however, are "major direct dischargers" as defined by law, including industrial direct
dischargers (facilities that discharge pollutants directly into water bodies) and Publicly Owned
Treatment Works (POTWs) (which receive and treat domestic and municipal waste and industrial
wastewater and discharge treated water into water bodies). Major direct dischargers are found in a
wide variety of industry sectors ranging from cement manufacturing to metal products and machinery
to petroleum refining. Major direct dischargers may be subject to industry-specific Effluent Limitation
Guidelines (ELGs),42 which are national technology-based standards for wastewater discharges to
surface waters (U.S. EPA, 2012b).
Technical details about the methodology used to determine proximity to major direct dischargers to
water are provided in Appendix C.
Indicator The count of major direct discharger facilities within 5 km, divided by
distance, presented as population-weighted averages of blocks in each
block group. Adjustments are made if there are none within 5 km, and so
that the minimum distance used is reasonable when very small.
Water pollutants can have human health or adverse ecological effects,
depending on concentration in the water, exposure to the water, toxicity
of the particular chemical and other factors. There are approximately
6,700 major direct dischargers in the United States. These facilities
discharge around 50 billion pounds of pollutants each year directly into the
nation's streams and rivers (including conventional pollutants such as
nitrogen and phosphorus) (U.S. EPA, 2012a).
People may be exposed to the discharged pollutants either directly or
through indirect pathways. People swimming in the downstream waters or
engaging in water-based recreation may be directly exposed dermally,
orally or through inhalation of volatized substances. If the released
substances reach a downstream drinking water intake, consumers of the
finished waters may consume whatever portion of the substances is not
removed by the drinking water utility. Some portion of the discharged
materials may enter the groundwater of neighboring areas and reach
42 A list of Effluent Guidelines by industry category can be found at
http://water.epa.gov/scitech/wastetech/guide/industrv.cfm.
Rationale for
Inclusion
55 | P a g e
-------
Details on Environmental Indicators
people through drinking water derived from wells that draw upon that
aquifer.
Some EPA EJ screening tools have included measures of proximity to all
facilities regulated by EPA, which includes major direct water dischargers.
More Information More information is at the water website (http://www.epa.gov/water) for
drinking water, rivers, and other categories (http://water.epa.gov/type/). a
local water quality webpage (http://water.epa.gov/drink/local/index.cfm).
and the NPDES webpage (http://water.epa.gov/polwaste/npdes/).
Relevant Studies Some examples of studies of disparities in proximity to water dischargers
include Fitos and Chakraborty (2010); Brender et al., 2011; VanDerslice,
2011; and a recent study (Deganian & Thompson, 2012) that tallied the
number of facilities of different types within 10 km2 squares and compared
these counts to the demographics of the squares.
Data Source Latitudes/Longitudes for major direct dischargers were taken from EPA's
PCS/ICIS database.
Data Version The 2015 version of EJSCREEN uses locational information retrieved from
the PCS/ICS database in May 2012 (U.S. EPA, 2012a). A total of 6,981 of the
major direct dischargers had suitable location information and were
included. Note that the point data used to show facility locations in the
"Map Supplementary Layers" menu is updated more often than the
database with calculated EJSCREEN indicators and indexes, so in some
small number of cases a facility may be in one data source but not the
other.
According to these data, almost 50% of the U.S. population lives in block
groups where at least one block's centroid is within 5 km of the nearest
NPDES facility. The population's mean score was 0.25, which could indicate
one facility at a distance of 4 km.
Discussion Monitored or modeled data on drinking or surface water quality could not
be identified with adequate national coverage and resolution. As more
data become available in the future, such data may be considered for
inclusion in EJSCREEN. As with all proximity-based indicators, proximity
alone may not represent any actual risk or even exposure.
Each block group in the United States was assigned a proximity score that
was the population-weighted sum of block-level proximity scores.
Appendix C provides more details on calculation of proximity scores. First,
each block was given a proximity score that was the sum of inverse
56 | P a g e
-------
Details on Environmental Indicators
distance-weighted count of facilities anywhere within 5 km of the block's
centroid. This score can be thought of as the number of facilities per
kilometer of distance from the average person. It is also equal to the
number of facilities divided by the harmonic mean of their distances. This
means one facility exactly 2 km away gives a score of 1/2, while three
facilities exactly 4 km away give a score of 3/4, and five facilities all at 1 km
away give a score of 5.43 If there is no facility within 5 km of a block
centroid, 1/d is used, where d is the distance to the single nearest at any
distance.
Proximity to major NPDES dischargers is only an indirect indicator of
potential exposure to water pollutants, because residents may not spend
time in or near the affected water, and the discharges may not impact local
drinking water supplies, or result in exposure from swimming or fishing.
43 An adjustment was made so that any distance smaller than 90% of the block's "radius" was set equal to 90% of
that radius, with radius defined as the square root of (area/pi). This adjustment accounted for the fact that the
average location (residence) within a circle of radius R is 0.9R away from the average point (facility) that is within
the circle. For more detail, see Appendix C.
57 | P a g e
-------
Details on Environmental Indicators
Proximity to NPL Sites
Congress enacted the Comprehensive Environmental Response, Compensation and Liability Act
(CERCLA), commonly known as Superfund, in 1980. This law was established to provide broad federal
authority to respond to uncontrolled abandoned hazardous waste sites. Under CERCLA, EPA's response
can involve remedial (long-term) cleanup actions or short-term removal actions.
EPA places sites on the National Priorities List (NPL) (a key subset of all "Superfund" sites) based on a
defined set of criteria and a public comment process. Inclusion of a site on the NPL does not impose a
financial obligation on EPA, nor does it assign liability to any party. The NPL serves primarily
informational purposes, identifying sites that appear to warrant remedial actions, thereby conveying to
policymakers and the public the size and nature of the nation's cleanup challenges.
Sites can be placed on the NPL in one of three ways44:
1. The site receives a score of 28.5 or higher in EPA's Hazard Ranking System (HRS);
2. States or territories designate a top-priority site; or
3. A site meets these requirements:
a. The Agency for Toxic Substances and Disease Registry (ATSDR) of the U.S. Public Health
Service has issued a health advisory that recommends removing people from the site;
b. EPA determines the site poses a significant threat to public health; and
c. EPA anticipates it will be more cost-effective to use its remedial authority (available only
at NPL sites) than to use its emergency removal authority to respond to the site.
Technical details about the methodology used to determine proximity to NPL sites are provided in
Appendix C.
Indicator The count of sites proposed and listed on the National Priorities List (NPL), each
represented by a point on the map (latitude/ longitude coordinate), within 5 km of
the average resident in a block group, divided by distance, calculated as the
population-weighted average of blocks in each block group. Adjustments are made
if there are no NPL sites within 5 km, and so that the minimum distance used is
reasonable when very small.
Soon after the passage of CERCLA and the Superfund Amendments and
Reauthorization Act, questions started to be raised about the locations, listing
decisions and pace of cleanup at NPL sites in low-income and minority communities
(Hird, 1993; Probst, 1990; United Church of Christ, 1987), and such concerns have
continued to this day (Anderton, Oakes, & Egan, 1997; Baden, Noonan, & Turaga,
2007; O'Neil, 2007). The study by Deganian & Thompson (2012) included NPL sites
in the tally of pollution points in each of the 10 km2 squares in the study area, for
comparison with demographic variables. Earlier studies related the presence of NPL
sites to population characteristics for different definitions of the host areas-
Rationale for
Inclusion
44 http://www.epa.gov/superfund/programs/npl hrs/nplon.htm
58 | P a g e
-------
Details on Environmental Indicators
counties (Hird, 1993), Census places or minor civil divisions where places are not
defined (Zimmerman, 1993), and Census tracts (Anderton et al., 1997).
The contaminants in NPL sites may reach humans in a number of ways. Volatile
contaminants may enter the atmosphere and reach individuals via the inhalation
route. Particularly in dry climates or seasons, contaminants on the surface of some
sites can become airborne and reach people directly through inhalation or
indirectly after being deposited on surfaces that people may contact. Contaminants
can also enter the food chain if the wind disperses them onto land used for
agriculture. Some contaminants may migrate into groundwater. People may be
exposed via drinking water derived from the aquifer, through vapor intrusion into
their residences or through other routes.
Some EPA EJ screening tools have included measures of proximity to all facilities or
other sites regulated by EPA, which include NPL sites.
More More information is available at the Superfund website
Information (http://www.epa.gov/superfund).
Relevant
Studies
Some examples of studies focused on disparities in proximity to NPL sites include
Brender et al. (2011) and those reviewed in Liu (2001).
59 | P a g e
-------
Details on Environmental Indicators
Data Source A single point location (latitude/ longitude coordinates) for each proposed and
listed NPL site was obtained from EPA's CERCLIS database. The database does not
provide details on the boundaries of each site, so this point data had to serve as a
way to represent site locations. For residents close to very large sites, the available
data may not provide an accurate representation of proximity to relevant portions
of the site. These points are approximations of the locations of sites, and are not
necessarily at the "center" of a given site. In a few cases a site's coordinates were
located in a major body of water according to the database, so EPA manually
specified new, plausible, nearby coordinates for use in EJSCREEN.
The count excludes deleted sites and sites in U.S. territories. Sites located in Guam
and Puerto Rico are not included in the 2015 version of EJSCREEN.
Data Version The 2015 version of EJSCREEN uses locational information retrieved from the
CERCLIS database in November 2013. A total of 1,350 proposed and listed NPL sites
were included in the EJSCREEN indicator. Note that the point data used to show site
locations in the "Map Supplementary Layers" menu is a different database than the
database used to calculate the EJSCREEN NPL proximity indicators and indexes. The
Superfund "Map Supplementary Layer" database includes deleted NPL sites, and
NPL sites in U.S. territories, excluded from the EJSCREEN NPL proximity indicator
database.
Discussion Each Census block group in the United States was assigned a proximity score that
was the population-weighted sum of block-level proximity scores. Appendix C
provides more details on how proximity scores were calculated. First, each Census
block was given a proximity score that was the sum of inverse distance-weighted
count of sites anywhere within 5 km of the block's internal point. This score can be
thought of as the number of NPL sites per kilometer of distance from the average
person. It is also equal to the number of sites divided by the harmonic mean of
their distances. This means one site 2 km away gives a score of 1/2, while three
sites each 4 km away give a score of 3/4, and five sites all at 1 km away give a score
of 5.45 If there is no site within 5 km of a block centroid, 1/d is used, where d is the
distance to the single nearest at any distance.
As with all proximity-based indicators, proximity alone may not represent any
actual risk or even exposure.
45 An adjustment was made so that any distance smaller than 90% of the block's "radius" was set equal to 90% of
that radius, with radius defined as the square root of (area/pi). This adjustment accounted for the fact that the
average location (residence) within a circle of radius R is 0.9R away from the average point (site) that is within the
circle. For more detail, see Appendix C.
60 | P a g e
-------
Details on Environmental Indicators
Proximity to TSDFs
The Resource Conservation and Recovery Act (RCRA), an amendment to the Solid Waste Disposal Act,
was enacted in 1976 to address the growing volumes of municipal and industrial solid waste generated
nationwide. RCRA was further amended in 1984 with the addition of the Hazardous and Solid Waste
Amendments. RCRA Subtitle C establishes a federal program to manage hazardous wastes from "cradle
to grave/' or from generation to disposal, to ensure that hazardous waste is managed in a manner that
protects human health and the environment. EPA has developed Subtitle C regulations governing
hazardous waste generation, transportation, and the several hundred active treatment, storage or
disposal facilities (TSDFs).46
Technical details about the methodology used to determine proximity to TSDF facilities are provided in
Appendix C.
Indicator The count of all commercial TSDF facilities within 5 km, divided by distance,
presented as population-weighted averages of blocks in each block group.
Adjustments are made if there are none within 5 km, and so that the
minimum distance used is reasonable when very small.
The substances at TSDF facilities may reach humans in a number of ways.
Volatile substances may enter the atmosphere and reach residents via the
inhalation route. Particularly in dry climates or seasons, substances on the
surface of some sites may be entrained in the atmosphere and reach people
directly through inhalation or indirectly after being deposited on surfaces
that people may contact or on arable land. Some substances may migrate
from the site into groundwater. People may be exposed via drinking water
derived from the aquifer, through vapor intrusion into their residences or
through other routes.
Some EPA EJ screening tools have included measures of proximity to all
facilities regulated by EPA, which includes TSDFs.
More Information More information is available at the hazardous waste webpage
(http://www.epa.gov/epawaste/hazard), the TSD webpage
(http://www.epa.gov/epawaste/hazard/tsd/). the waste information page
(http://www.epa.gov/epawaste/inforesources/) and a RCRAInfo page
(http://www.epa.gov/epawaste/inforesources/data).
46 Basic information and TSDF counts are available here:
http://www.epa.gov/compliance/data/results/performance/rcra/index.html
More information is available from EPA's RCRA Orientation Manual, available at
http://www.epa.gov/osw/inforesources/pubs/orientat/ (U.S. EPA, 2012e).
Rationale for
Inclusion
61 | P a g e
-------
Details on Environmental Indicators
Relevant Studies Some examples of studies or reviews that have focused on disparities in
proximity to TSDFs include Liu (2001) and Brender et al., (2011). Issues
around environmental justice and TSDFs influenced the early origins of EJ
work (General Accounting Office, 1983; United Church of Christ, 1987) and
have been the topic of ongoing research (Been & Gupta, 1997; Boer, Pastor
Jr., Sadd, & Synder, 1997; Mohai & Saha, 2007; Oakes, Anderton, &
Anderson, 1996; Pastor Jr., Sadd, & Hipp, 2001; Saha & Mohai, 2005; United
Church of Christ, 2007).
The study by Deganian & Thompson (2012) included Hazardous Waste
Inventory sites, RCRA hazardous waste storage sites and active solid waste
landfills sites in the tally of pollution points in each of the 10 km2 squares in
the study area, for comparison with demographic variables.
Earlier studies related the presence of TSDFs to population characteristics for
different definitions of the host areasCensus-designated areas (General
Accounting Office, 1983), postal ZIP codes (United Church of Christ, 1987),
and Census tracts (Anderton, Anderson, Oakes, & Fraser, 1994).
Data Source Latitudes/Longitudes for all active commercial TSDF sites were obtained
from the RCRAInfo database.
Data Version The 2015 version of EJSCREEN uses locational information retrieved from the
RCRAInfo database in May 2012 (U.S. EPA, 2011e). A total of 586 TSDF
facilities were included in this version of EJSCREEN. Note that the point data
used to show facility locations in the "Map Supplementary Layers" menu is
updated more often than the database with calculated EJSCREEN indicators
and indexes, so in some small number of cases a facility may be in one data
source but not the other.
Discussion Each block group in the United States was assigned a proximity score that
was the population-weighted sum of block-level proximity scores. Appendix
C provides more details on the calculation of proximity scores. First, each
block was given a proximity score that was the sum of inverse distance-
weighted count of TSDFs anywhere within 5 km of the block's centroid. This
score can be thought of as the number of facilities per kilometer of distance
from the average person. It is also equal to the number of facilities divided
by the harmonic mean of their distances. This means one facility exactly 2
km away gives a score of 1/2, while three facilities exactly 4 km away give a
score of 3/4, and five facilities all at 1 km away give a score of 5.47 If there is
47 An adjustment was made so that any distance smaller than 90% of the block's "radius" was set equal to 90% of
that radius, with radius defined as the square root of (area/pi). This adjustment accounted for the fact that the
62 | P a g e
-------
Details on Environmental Indicators
no facility within 5 km of a block centroid, 1/d is used, where d is the
distance to the single nearest at any distance.
As with all proximity-based indicators, proximity alone may not represent
any actual risk or even exposure.
average location (residence) within a circle of radius R is 0.9R away from the average point (facility) that is within
the circle. For more detail, see Appendix C.
63 | P a g e
-------
Details on Environmental Indicators
Proximity to RMP Sites
Accidental releases of toxic substances and incidents involving fires and explosions can result from the
production, use, or transport of industrial materials. Evacuations, injuries and deaths have resulted in
some cases. Concern about the risks of chemical accidents led Congress to pass the Emergency Planning
and Community Right-to-Know Act of 1986 (EPCRA), and amendments to the Clean Air Act (CAA)
(section 112(r)), which together created reporting and planning obligations for a variety of facility types,
the EPA, and state and local planning and response organizations.
The facilities discussed here as "RMP facilities" are those facilities required by the CAA to file risk
management plans. The regulations under CAA section 112(r) establishes a List of Regulated
Substances72 substances listed because of their high acute toxicity and 60 because of their flammable
or explosive potentialalong with threshold quantities (TQs) for each. The listed substances are those
that pose the greatest risk of harm from accidental releases. If a facility maintains a quantity of any such
chemical above those TQs, it must file an RMP with EPA.
It should be noted that some concerns related to proximity to facilities are already accounted for in
NATA indicators for ambient air pollutants (e.g., cancer risk and hazard indexes), but NATA is based on
one year of reported annual releases, which would not account for accidental releases unless they
occurred that year.
Technical details about the methodology used to determine proximity to RMP facilities are provided in
Appendix C.
Indicator The count of RMP facilities within 5 km, divided by distance, presented as
population-weighted averages of blocks in each block group. Adjustments
are made if there are none within 5 km, and so that the minimum distance
used is reasonable when very small.
Rationale for RMP facilities are diverse in their size, structure, activities and the makeup
Inclusion of the regulated substances. As with many types of industrial facilities,
there may be routine releases to the air and water of the residuals after
pollution control devices remove what is generally a large fraction of the
waste stream. Thus, people may be exposed to some substances directly
through inhalation or indirectly through water routes or via ingestion of
food. But the primary concerns with RMP facilities are the accidental
release of substances and fires or explosions. The sudden release of
relatively large quantities of acutely toxic substances can cause serious
health effects including death after inhalation or dermal exposure. These
effects may be prompt or may occur or persist for some time after
exposure. Fires may affect neighboring areas and the associated smoke
may expose people to toxic combustion products. Explosions may cause
material damage and injuries to people in neighboring areas. Local
64 | P a g e
-------
Details on Environmental Indicators
residents, as well as workers and emergency responders, may suffer severe
adverse effects.
Some EPA EJ screening tools have included measures of proximity to all
facilities regulated by EPA, which include RMP facilities.
More Information More information is available at the RMP program webpage
(http://www2.epa.gov/rmp) and the RMP Info database stored in
Envirofacts (http://www.epa.gov/enviro/facts/rcrainfo/search.html).
Relevant Studies The EJ literature contains numerous studies that have examined proximity
to various types of sites, including some relevant to the possibility or
frequency of chemical accidents. Since the 1980s, many studies have
examined the frequency and consequences of accidental releases of
acutely toxic chemicals or events resulting in fires or explosions (Binder,
1989, and many other studies). After the RMP program was established,
researchers examined the characteristics of the RMP reporting facilities
and their reported accident histories for insights into causes,
consequences, prevention and emergency response (Kleindorfer et al.,
2003).
Fewer studies have focused specifically on the relationship of RMP facilities
to the demographics of the surrounding populations. Disparity in acute
exposures to hazardous substances was addressed by Chakraborty (2001).
The study by Deganian & Thompson (2012) included two categories of
facilitiesfacilities having air pollution permits and facilities reporting to
the Toxics Release Inventory programwhich include some overlap with
RMP facilities, but are not limited to RMP facilities. M. R. Elliott, Wang,
Lowe, & Kleindorfer (2004) examined the characteristics of RMP-reporting
facilities and their reported 5-year accident history versus the demographic
characteristics of the counties in which they are located. The demographic
characteristics examined included total population, race, education and
income. The study found an association between the presence of larger
and more chemical-intensive facilities and counties with larger African-
American populations, and in counties with high levels of income
inequality but higher median incomes. Further, the study found a greater
risk of accidents for facilities in heavily African-American counties.
Data Source Latitudes/Longitudes for RMP facilities were obtained from EPA's RMP
database.
Data Version The 2015 version of EJSCREEN uses locational information retrieved from
the RMP database in May 2012. A total of 12,759 RMP facilities were
65 | P a g e
-------
Details on Environmental Indicators
included in the proximity indicators and related EJ indexes in this version of
EJSCREEN. Note that the point data used to show facility locations in the
"Map Supplementary Layers" menu is updated more often than the
database with calculated EJSCREEN indicators and indexes, so in some
small number of cases a facility may be in one data source but not the
other.
Discussion Each block group in the United States was assigned a proximity score that
was the population-weighted sum of block-level proximity scores.
Appendix C provides more details on the calculation of proximity scores.
First, each block was given a proximity score that was the sum of inverse
distance-weighted count of facilities anywhere within 5 km of the block's
internal point. This score can be thought of as the number of RMP facilities
per kilometer of distance from the average person. It is also equal to the
number of facilities divided by the harmonic mean of their distances. This
means one facility exactly 2 km away gives a score of 1/2, while three
facilities exactly 4 km away give a score of 3/4, and five facilities all at 1 km
away give a score of 5.48 If there is no facility within 5 km of a block
centroid, 1/d is used, where d is the distance to the single nearest at any
distance.
As with all proximity-based indicators, proximity alone may not represent
any actual risk or even exposure.
48 An adjustment was made so that any distance smaller than 90% of the block's "radius" was set equal to 90% of
that radius, with radius defined as the square root of (area/pi). This adjustment accounted for the fact that the
average location (residence) within a circle of radius R is 0.9R away from the average point (facility) that is within
the circle. For more detail, see Appendix C.
66 | P a g e
-------
4 BIBLIOGRAPHY
Anderton, D. L., Anderson, A. B., Oakes, J. M., & Fraser, M. R. (1994). Environmental equity: The demographics
of dumping. Demography, 31(2), 229-248.
Anderton, D. L., Oakes, J. M., & Egan, K. L. (1997). Environmental equity in Superfund. Demographics of the
discovery and prioritization of abandoned toxic sites. Eval Rev, 21(1), 3-26.
Baden, B. M., Noonan, D. S., & Turaga, R. M. R. (2007). Scales of justice: Is there a geographic bias in
environmental equity analysis? J Environ Planning Manage, 50(2), 163-185.
Baumann, L. M., Robinson, C. L., Combe, J. M., Gomez, A., Romero, K., Gilman, R. H., . . . Checkley, W. (2011).
Effects of distance from a heavily transited avenue on asthma and atopy in a periurban shantytown in
Lima, Peru .J Allergy Clin Immunol, 127(4), 875-882.
Been, V., & Gupta, F. (1997). Coming to the Nuisance or Going to the Barrios-A Longitudinal Analysis of
Environmental Justice Claims. Ecol Law Q, 24, 1-56.
Bell, M. L., & Ebisu, K. (2012). Environmental Inequality in Exposures to Airborne Particulate Matter
Components in the United States. Environ Health Perspect, 120(12), 1699-1704.
Bell, M. L., Zanobetti, A., & Dominici, F. (2014). Who is more affected by ozone pollution? A systematic review
and meta-analysis. Am J Epidemiol, 180(1), 15-28. http://dx.doi.org/10.1093/aje/kwull5
Berrocal, V. J., Gelfand, A. E., & Holland, D. M. (2010a). A bivariate space-time downscaler under space and
time misalignment. Ann Appl Stat, 4(4), 1942-1975.
Berrocal, V. J., Gelfand, A. E., & Holland, D. M. (2010b). A Spatio-Temporal Downscaler for Output From
Numerical Models. J Agric Biol Environ Stat, 15(2), 176-197.
Berrocal, V. J., Gelfand, A. E., & Holland, D. M. (2011). Space-Time Data fusion Under Error in Computer Model
Output: An Application to Modeling Air Quality. Biometrics, First published online: 29 Dec 2011.
Binder, S. (1989). Deaths, injuries, and evacuations from acute hazardous materials releases. Am J Public
Health, 79(8), 1042-1044.
67 | P a g e
-------
Boer, J. T., Pastor Jr., M., Sadd, L. L., & Synder, L. D. (1997). Is There Environmental Racism? The Demographics
of Hazardous Waste in Los Angeles County. Soc Sci Quart, 78(4), 793-810.
Bravo, M. A., Fuentes, M., Zhang, Y., Burr, M. J., & Bell, M. L. (2012). Comparison of exposure estimation
methods for air pollutants: Ambient monitoring data and regional air quality simulation. Environ Res,
116(1-10).
Brender, J. D., Maantay, J. A., & Chakraborty, J. (2011). Residential Proximity to Environmental Hazards and
Adverse Health Outcomes. Am J Public Health, 101(SI), S19-S26.
Briguglio, L. (1997). Alternative Economic Vulnerability Indices for Developing Countries. Report prepared for
the Expert Group on Vulnerability Index. UN (DESA), 17-19, December.
Brochu, P. J., Yanosky, J. D., Paciorek, C. J., Schwartz, J., Chen, J. T., Herrick, R. F., & Suh, H. H. (2011). Particulate
air pollution and socioeconomic position in rural and urban areas of the Northeastern United States.
Am J Public Health, 101(S1), S224-S230.
Byun, D., & Schere, K. (2006). Review of the governing equations, computational algorithms, and other
components of the models-3 Community Multiscale Air Quality (CMAQ) modeling system. Appl Mech
Rev, 59, 51-77.
Cauthen, N. K., & Fass, S. (2008). Measuring Poverty in the United States: National Center for Children in
Poverty. Mailman School of Public Health.
Centers for Disease Control and Prevention. (2010). Behavioral Risk Factor Surveillance System Survey Data.
Atlanta: U.S. Department of Health and Human Services, http://www.cdc.gov/brfss/
Centers for Disease Control and Prevention. (2011a). CDC Health Disparities and Inequalities Report United
States, 2011. MMWR, January 14, 2011; Vol. 60 (Suppl). Retrieved from
http://www.cdc.gov/mmwr/pdf/other/su6001.pdf.
Centers for Disease Control and Prevention. (2011b). Deaths: Leading Causes for 2007. National Vital Statistics
Reports. Vol. 59, No. 8. Retrieved from http://www.cdc.gov/nchs/data/nvsr/nvsr59/nvsr59 08.pdf.
Centers for Disease Control and Prevention. (2011c). Epi Info. Retrieved April 12, 2012, from
http://wwwn.cdc.gov/epiinfo/.
68 | P a g e
-------
Centers for Disease Control and Prevention. (2011d). MMWR, September 9, 2011; Vol. 60, No. 35. Retrieved
from http://www.cdc.gov/mmwr/pdf/wk/mm6035.pdf.
Centers for Disease Control and Prevention. (2012). CDC Response to Advisory Committee on Childhood Lead
Poisoning Prevention Recommendations in "Low Level Lead Exposure Harms Children: A Renewed Call of
Primary Prevention". Retrieved from
http://www.cdc.gov/nceh/lead/acclpp/cdc response lead exposure recs.pdf.
Chakraborty, J. (2001). Acute Exposure to Extremely Hazardous Substances: An Analysis of Environmental
Equity. Risk Analysis, 21(5), 883-894.
Chakraborty, J. (2006). Evaluating the Environmental Justice Impacts of Transportation Improvement Projects in
the US. Transp Res D Transp Environ, 11(5), 315-323.
Chakraborty, J., Maantay, J. A., and Brender, J. D. (2011). Disproportionate Proximity to Environmental Health
Hazards: Methods, Models, and Measurement. Am J Public Health, 101 (SI), S27-S36.
Clark, L. P., Millet, D. B., & Marshall, J.D. (2014). National Patterns in Environmental Injustice and Inequality:
Outdoor N02 Air Pollution in the United States. PLoS ONE 9(4): e94431.
http://dx.doi.10.1371/iournal.pone.0094431
Cohen, R. A., & Martinez, M. E. (2011). Health Insurance Coverage: Early Release of Estimates From the
National Health Interview Survey, January-March 2011: National Center for Health Statistics.
deFur, P. L., Evans, G. W., Cohen Hubal, E. A., Kyle, A. D., Morello-Frosch, R. A., & Williams, D. R. (2007).
Vulnerability as a Function of Individual and Group Resources in Cumulative Risk Assessment. Environ
Health Perspect, 115(5), 817-824.
Deganian, D., & Thompson, J. (2012). The Patterns of Pollution: A Report on Demographics and Pollution in
Metro Atlanta: GreenLaw.
Elliott, M. R., Wang, Y., Lowe, R. A., & Kleindorfer, P. R. (2004). Environmental justice: Frequency and severity of
US chemical industry accidents and the socioeconomic status of surrounding communities. J Epidemiol
Community Health, 58, 24-30.
Environmental Health Sciences. Environmental Health News. Online periodical. Retrieved August 24, 2012,
from www.environmentalhealthnews.org.
69 | P a g e
-------
Fann, N., Lamson, A. D., Anenberg, S. C., Wesson, K., Risley, D., & Hubbell, B. J. (2012). Estimating the national
public health burden associated with exposure to ambient PM2.5 and ozone. Risk Analysis, 32(1), 81-
95.
Fann, N., & Risley, D. (2011). The public health context for PM2.5 and ozone air quality trends. Air Qual Atmos
Health, [Epub ahead of print January 5, 2011],
Fann, N., Roman, H. A., Fulcher, C. M., Gentile, M. A., Hubbell, B. J., Wesson, K., & Levy, J. I. (2011). Maximizing
health benefits and minimizing inequality: Incorporating local-scale data in the design and evaluation of
air quality policies. Risk Analysis, 31(6), 908-922.
Fedstats. (2007). FedStats. Retrieved April 12, 2012, from http://www.fedstats.gov/.
Finkel, A. M., & Golding, D. (Eds.). (1994). Worst Things First? The Debate Over Risk-Based National
Environmental Priorities. Washington DC: Resources for the Future.
Fitos, E., and Chakraborty, J. (2010). Race, Class, and Wastewater Pollution. In J. Chakraborty and M. M.
Bosman (Eds.) Spatial and Environmental Injustice in an American Metropolis: A Study of Tampa Bay,
Florida. Cambria Press: Amherst, NY; 59-82.
Fujita, E. M., Campbell, D. E., Zielinska, B., Arnott, W. P., & Chow, J. C. (2011). Concentrations of air toxics in
motor vehicle-dominated environments, Research Report 156. Boston: Health Effects Institute.
Gaitens, J. M., Dixon, S. L., Jacobs, D. E., Nagaraja, J., Wilson, J. W., & Ashley, P. J. (2009). Exposure of U.S.
children to residential dust lead, 1999-2004: I. Housing and demographic factors. Environ Health
Perspect, 117(3), 461-467.
Galea, S., Tracy, M., & Hoggatt, K. J. (2011). Estimated Deaths Attributable to Social Factors in the United
States. Am J Public Health, 101(8), 1456-1465.
General Accounting Office. (1983). Siting of Hazardous Waste Landfills and their Correlation with Racial and
Economic Status of Surrounding Communities. Washington, DC: Retrieved from
http://www.gao.gov/products/121648.
Grineski, S. (2007). Incorporating Health Outcomes into Environmental Justice Research: The Case of Children's
Asthma and Air Pollution in Phoenix, Arizona. Environmental Hazards, 7, 360-371.
70 | P a g e
-------
Grineski, S., Bolin, B., and Boone, C. (2007). Criteria Pollution and Marginal Populations: Environmental Inequity
in metropolitan Phoenix, Arizona, USA. SocSci Q, 88, 535-554.
Health Effects Institute. (2010). Traffic-Related Air Pollution: A Critical Review of the Literature on Emissions,
Exposure, and Health Effects (HEI Special Report 17).
Hird, J. A. (1993). Environmental policy and equity: The case of superfund. J Policy Anal Manag, 12(2), 323-343.
Hoek, G., Beelen, R., Kos, G., Dijkema, M., Zee, S. C. v. d., Fischer, P. H., & Brunekreef, B. (2011). Land Use
Regression Model for Ultrafine Particles in Amsterdam. Environ Sci Technol, 45(2), 622-628.
Hoffman, B., Moebus, S., Dragano, N., Mohlenkamp, S., Memmesheimer, M., Erbel, R., & Jockel, K. H. (2009).
Residential traffic exposure and coronary heart disease: results from the Heinz Nixdorf Recall Study.
Biomarkers, 14(SI), 74-78.
Hystad, P., Setton, E., Cervantes, A., Poplawski, K., Deschenes, S., Brauer, M., . .. Demers, P. (2011). Creating
national air pollution models for population exposure assessment in Canada. Environ Health Perspect,
119(8), 1123-1129.
Jacobs, D. E., Clickner, R. P., Zhou, J. Y., Viet, S. M., Marker, D. A., Rogers, J. W., .. . Friedman, W. (2002). The
Prevalence of Lead-Based Paint Hazards in U.S. Housing. Environ Health Perspect, 110(10), a599-a606.
Jones, R. L., Homa, D. M., Meyer, P. A., Brody, D. J., Caldwell, K. L., Pirkle, J. L., & Brown, M. J. (2009). Trends in
blood lead levels and blood lead testing among US children aged 1 to 5 years, 1988-2004. Pediatrics,
123(3), e376-e385.
Kleindorfer, P. R., Belke, J. C., Elliott, M. R., Lee, K., Lowe, R. A., & Feldman, H. I. (2003). Accident epidemiology
and the U.S. chemical industry: Accident history and worst-case data from RMP*lnfo. Risk Analysis,
23(5), 865-881.
Krewski, D., Jerrett, M., Burnett, R. T., Ma, R., Hughes, E., Shi, Y., . . . Thun, M. J. (2009). Extended Follow-Up
and Spatial Analysis of the American Cancer Society Study Linking Particulate Air Pollution and
Mortality, Research Report 140. Boston: Health Effects Institute.
Levin, R., Brown, M. J., Kashtock, M. E., Jacobs, D. E., Whelan, E. A., Rodman, J., . . . Sinks, T. (2008). Lead
Exposures in U.S. Children, 2008: Implications for Prevention. Environ Health Perspect, 116(10), 1285-
1293.
71 | P a g e
-------
Levy, J. I., Wilson, A. M., & Zwack, L. M. (2007). Quantifying the Efficiency and Equity Implications of Power
Plant Air Pollution Control Strategies in the United States. Environ Health Perspect, 115(5), 743-750.
Liu, F. (2001). Environmental Justice Analysis: Theories, Methods, and Practice. Boca Raton: CRC Press.
https://www.crcpress.com/product/isbn/9781566704038
Lobdell, D. T., Jagai, J. S., Rappazzo, K., & Messer, L. C. (2011). Data sources for an environmental quality index:
Availability, quality, and utility. Am J Public Health, 101(S1), S277-S285.
Maantay, J., Chakraborty, J., & Brender, J. (2010). Proximity to Environmental Hazards: Environmental Justice
and Adverse Health Outcomes, Final Revised Draft. Prepared for the U.S. Environmental Protection
Agency "Strengthening Environmental Justice Research and Decision Making: A Symposium on the
Science of Disproportionate Environmental Health Impacts". Final publication available at
http://dx.doi.org/10.2105/AJPH.2011.30Q183
Maroko, A. R. (2012). Using air dispersion modeling and proximity analysis to assess chronic exposure to fine
particulate matter and environmental justice in New York City. Appl Geogr, 34, 533-547.
McMillan, N. J., Holland, D. M., Morara, M., & Feng, J. (2010). Combining numerical model output and
particulate data using Bayesian space-time modeling. Environmetrics, 21(1), 48-65.
Messer, L. C., Jagai, J. S., Rappazzo, K. M., Lobdell, D. T. (2014). Construction of an environmental quality index
for public health research. Environ Health. May 22;13(1):39. http://dx.doi.org/10.1186%2F1476-069X-
13-39
Miranda, M. L., Edwards, S. E., Keating, M. H., & Paul, C. J. (2011). Making the Environmental Justice Grade: The
Relative Burden of Air Pollution Exposure in the United States. IntJ Environ Res Public Health, 8(6).
Mohai, P., & Saha, R. (2007). Racial Inequality in the Distribution of Hazardous Waste: A National-Level
Reassessment. Social Problems, 54(3), 343-270.
Morello-Frosch, R., & Jesdale, B. M. (2006). Separate and Unequal: Residential Segregation and Estimated
Cancer Risks Associated with Ambient Air Toxics in U.S. Metropolitan Areas. Environ Health Perspect,
114(3), 386-393.
National Research Council of the National Academies Committee on Improving Risk Analysis Approaches Used
by the U.S. EPA. (2009). Science and Decisions: Advancing Risk Assessment. Washington, DC: The
National Academies Press, http://www.nap.edu/catalog/12209/science-and-decisions-advancing-risk-
assessment
72 | P a g e
-------
National Environmental Justice Advisory Council (NEJAC). (2010). Recommendations for Nationally Consistent
Environmental Justice Screening Approaches.
http://www.epa.gov/environmentaliustice/resources/publications/neiac/ei-screening-approaches-rpt-
2010.pdf via http://www.epa.gov/environmentaliustice/neiac/recommendations.html
O'Neil, S. G. (2007). Superfund: Evaluating the Impact of Executive Order 12898. Environ Health Perspect,
115(7), 1087-1093.
Oakes, J. M., Anderton, D. L., & Anderson, A. B. (1996). A Longitudinal Analysis of Environmental Equity in
Communities with Hazardous Waste Facilities. SocSciRes, 25, 125-148.
The Organisation for Economic Co-operation and Development (OECD). (2008). Handbook on Constructing
Composite Indicators.
http://www.oecd.org/els/soc/handbookonconstructingcompositeindicatorsmethodologyanduserguide.
htm
Pastor Jr., M., Morello-Frosch, R., & Sadd, J. (2010). Final Report: Contract Number 04-308. Air Pollution and
Environmental Justice: Integrating Indicators of Cumulative Impact and Socio-Economic Vulnerability
into Regulatory Decision-Making.
Pastor Jr., M., Sadd, J., & Hipp, J. (2001). Which Came First? Toxic Facilities, Minority Move-In, and
Environmental Justice. J Urban Affairs, 23(1), 1-21.
Post, E. S., Belova, A., & Huang, J. (2011). Distributional Benefit Analysis of a National Air Quality Rule. IntJ
Environ Res Public Health, 8(6), 1872-1892.
Probst, K. N. (1990). Hazardous Waste and the Rural Poor: A Preliminary Assessment: Clean Sites, Inc.
Rosenbaum, A., Hartley, S., & Holder, C. (2011). Analysis of diesel particulate matter health risk disparities in
selected US harbor areas. Am J Public Health, 101(SI), S217-S223.
Rowangould, G.M. (2013). A census of US near-roadway population: Public health and environmental justice
considerations. Transp Res D Transp Environ, 25, 59-67.
Sadd, J. L., Pastor, M., Morello-Frosch, R., Scoggins, J., & Jesdale, B. (2011). Playing it safe: Assessing cumulative
impact and social vulnerability through an environmental justice screening method in the South Coast
Air Basin, California. IntJ Environ Res Public Health, 8(5), 1441-1459.
73 | P a g e
-------
Saha, R., & Mohai, P. (2005). Historical Context and Hazardous Waste Facility Siting: Understanding Temporal
Patterns in Michigan. Social Problems, 52(4), 618-648.
S0rensen, M., Hvidberg, M., Andersen, Z. J., Nordsborg, R. B., Lillelund, K. G., Jakobsen, J., . . . Raaschou-Nielsen,
O. (2011). Road traffic noise and stroke: A prospective cohort study. Eur Heart J, 32(6), 737-744.
Spengler, J., Lwebuga-Mukasa, J., Vallarino, J., Melly, S., Chillrud, S., Baker, J., & Minegishi, T. (2011). Air Toxics
Exposure from Vehicle Emissions at a U.S. Border Crossing: Buffalo Peace Bridge Study, Research
Report 158. Boston: Health Effects Institute.
Su, J. G., Morello-Frosch, R., Jesdale, B. M., Kyle, A. D., Shamasunder, B., & Jerrett, M. (2009). An index for
assessing demographic inequalities in cumulative environmental hazards with application to Los
Angeles, California. Environ Sci Technol, 43(20), 7626-7634.
Tian, N., Xue, J., & Barzyk, T.M. (2013). Evaluating socioeconomic and racial differences in traffic-related
metrics in the United States using a GIS approach. J Exposure Science Envt Epidem, 23, 215-222.
U.S. Census Bureau. (2013). American Community Survey, 5-year Estimates 2008-2012. Data retrieved in 2013,
from http://www2.census.gov/acs2012 5yr/summaryfile/ and
http://www.census.gov/acs/www/data documentation/summary file/
U.S. Department of Transportation. (2011). National Transportation Atlas Database. Highway Performance
Monitoring System. Data retrieved April 12, 2012, from
http://www.rita.dot.gov/bts/sites/rita.dot.gov.bts/files/publications/national transportation atlas dat
abase/2011/index.html.
U.S. EPA. (1987). Unfinished Business: A Comparative Assessment of Environmental Problems. Washington, DC:
Retrieved from http://nepis.epa.gov/Exe/ZvPURL.cgi?Dockev=2000BZQP.txt.
U.S. EPA. (1992). Environmental Equity: Reducing the Risk for All Communities. Washington, DC: Retrieved from
http://www.epa.gov/compliance/ei/resources/reports/annual-proiect-
reports/reducing risk com voll.pdf.
U.S. EPA. (2006a). Air Quality Criteria for Lead (2006) Final Report. EPA/600/R-05/144aF-bF. Washington, DC.
http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=158823
U.S. EPA. (2006b). Air Quality Criteria for Ozone and Related Photochemical Oxidants. Washington, DC:
Retrieved from http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=149923.
74 | P a g e
-------
U.S. EPA. (2009a). A Framework for Categorizing the Relative Vulnerability of Threatened and Endangered
Species to Climate Change (External Review Draft). EPA/600/R-09/011. Washington, DC.
http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm7deich203743
U.S. EPA. (2009b). Integrated Science Assessment for Particulate Matter (Final Report). Washington, DC:
Retrieved from http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=216546.
U.S. EPA. (2009c). National-Scale Air Toxics Assessment for 2002 - Fact Sheet. Retrieved April 12, 2012, from
http://www.epa.gov/ttn/atw/nata2002/factsheet.html.
U.S. EPA. (2009d). Original list of hazardous air pollutants. Retrieved April 12, 2012, from
http://www.epa.gov/ttnatw01/187polls.html.
U.S. EPA. (2010). Interim Guidance on Considering Environmental Justice During the Development of an Action.
Retrieved from http://www.epa.gov/compliance/ei/resources/policv/considering-ei-in-rulemaking-
guide-07-2010.pdf.
U.S. EPA. (2011a). Exposure Factors Handbook: 2011 Edition. Washington, DC: Retrieved from
http://www.epa.gov/ncea/efh/pdfs/efh-complete.pdf.
U.S. EPA. (2011b). [reference no longer used],
U.S. EPA. (2011c). Integrated Science Assessment for Lead (1st External Review Draft 2011 & Final Report 2013).
Washington, DC. EPA/600/R-1 0/075F. http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=255721
U.S. EPA. (2011d). [reference no longer used]
U.S. EPA. (2011e). RCRAInfo database search in Envirofacts. Data retrieved April 12, 2012, from
http://www.epa.gov/enviro/facts/rcrainfo/search.html.
U.S. EPA. (2012a). Facilities and Enforcement Activities related to the Clean Water Act National Pollutant
Discharge Elimination System (NPDES) Program. Data retrieved April 12, 2012, from
http://www.epa.gov/oecaerth/data/results/performance/cwa/index.html.
U.S. EPA. (2012b). Industrial Regulations. Retrieved April 12, 2012, from
http://water.epa.gov/scitech/wastetech/guide/industry.cfm.
75 | 3 a g e
-------
U.S. EPA. (2012c). Integrated Risk Information System (IRIS). Retrieved April 12, 2012, from
http://www.epa.gov/IRIS/.
U.S. EPA. (2012d). Our Nation's Air: Status and Trends Through 2010. Research Triangle Park: Retrieved from
http://www.epa.gov/airtrends/2011/report/fullreport.pdf.
U.S. EPA. (2012e). RCRA Orientation Manual 2011: Resource Conservation and Recovery Act. Retrieved August
24, 2012, from http://www.epa.gov/osw/inforesources/pubs/orientat/.
U.S. EPA. (2012f). Search Superfund Site Information. Data retrieved May, 2012, from
http://cumulis.epa.gov/supercpad/cursites/srchsites.cfm.
U.S. EPA. (2012g). Summary Nonattainment Area Population Exposure Report. Retrieved August 24, 2012, from
http://www.epa.gov/oar/oaqps/greenbk/popexp.html.
U.S. EPA. (2015a). Fused Air Quality Surfaces Using Downscaling. Data retrieved from
http://www.epa.gov/esd/land-sci/lcb/lcb faqsd.html.
U.S. EPA. (2015b). National-Scale Air Toxics Assessments. Data retrieved from
http://www.epa.gov/ttn/atw/natamain/index.html.
United Church of Christ. (1987). Toxic Waste and Race in the United States: A National Report on the Racial and
Socio-Economic Characteristics of Communities with Hazardous Waste Sites: Commission for Racial
Justice, http://www.ucc.org/environmental-ministries toxic-waste-20
United Church of Christ. (2007). Toxic Wastes and Race at Twenty: 1987-2007. Cleveland: Justice and Witness
Ministries, http://www.ucc.org/environmental-ministries toxic-waste-20
van Donkelaar, A., Martin, R. V., Brauer, M., Kahn, R., Levy, R., Verduzco, C., & Villeneuve, P. J. (2010). Global
Estimates of Ambient Fine Particulate Matter Concentrations from Satellite-Based Aerosol Optical
Depth: Development and Application. Environ Health Perspect, 118(6), 847-855.
http://dx.doi.org/10.1289/ehp.09Q1623
VanDerslice, J. (2011). Drinking water infrastructure and environmental disparities: evidence and
methodological considerations. Am J Public Health, 101(S1), S109-S114.
76 | P a g e
-------
Villagran de Leon, J. C. (2006). Vulnerability: A Conceptual and Methodological Review (Vol. 4/2006): SOURCE
(Studies Of the University: Research, Counsel, Education), United Nations University - Institute for
Environmental and Human Security.
Yang, T.-C., & Matthews, S. A. (2010). The Role of Social and Built Environments in Predicting Self-rated Stress:
A Multilevel Analysis in Philadelphia. Health & Place, 16(5), 803-810.
Zhou, Y., & Levy, J. I. (2007). Factors influencing the spatial extent of mobile source air pollution impacts: A
meta-analysis. BMC Public Health, 7(89).
Zhu, Y., Hinds, W. C., Kim, S., & Sioutas, C. (2002). Concentration and size distribution of ultrafine particles near
a major highway. J Air Waste Manag Assoc, 52(9), 1032-1042.
Zimmerman, R. (1993). Social Equity and Environmental Risk. Risk Analysis, 13(6), 649-666.
77 | P a g e
-------
Appendix A
APPENDIX A. DEVELOPMENT OF EJSCREEN
Review of Available Data and Other Tools
Preliminary planning for EJSCREEN began in late 2010. The first steps involved a review of existing or planned EJ
screening methods from EPA program and regional offices. EJSCREEN draws upon a great deal of prior research,
analysis and public involvement in the development of very closely related screening efforts. Early steps also
included a review of current EJ research on methods and data for EJ analysis. Information was gathered from
the following sources, among others:
Stakeholder and expert presentations at EPA's March 2010 conference on environmental justice.
EPA's 2010 expert workshop on economics and environmental justice.
EPA's ORD's C-FERST research program, including a review of data sources for EJ analysis.
EPA's ORD's Environmental Quality Index (EQI) compilation and review of data sources for
environmental indicators.
A review of several national reviews of analytic methods including the use of inequality metrics
(e.g. as presented in the expert workshop).
Review of EPA guidance documents and related documents on environmental justice49
Review of the NEJAC report of May 2010 on EJ screening (NEJAC, 2010).
Review of prior tools including EJSEAT, EJVIEW, various EPA Regional tools, and some state EJ
screening tools such as CalEnviroScreen.
Selecting an Approach to EJ Screening
A number of important considerations must be balanced when selecting an approach to an EJ screening tool:
Useful to end-users and other stakeholders.
Reflects EPA policies and EJ policy goals.
Reflects sound science.
Is feasible to develop and maintain, update and upgrade.
Data coverage and quality considerations are also discussed in the chapters describing the environmental
indicators.
EJSCREEN was developed through an EPA workgroup with participation from a very wide range of program
offices and Regional offices, and in consultation with management and scientists representing the various
offices, building upon the public input and scientific information developed in the course of prior screening
efforts such as EJSEAT and Regional experience with EJ screening. Quality control and peer review of EJSCREEN
are described in Appendices F and G.
[49 http://www.epa.gov/environmentaliustice/resources/policv/index.html
78 | P a g e
-------
Appendix B
[intentionally blank]
79 | P a g e
-------
Appendix B
APPENDIX B. TECHNICAL DETAILS ON PERCENTILES, ROUNDING,
BUFFERING, AND DEMOGRAPHIC DATA
How Percentiles are Calculated and Displayed
Percentiles, such as "80th percentile," are displayed in EJSCREEN as rounded down to the closest percentile
that is lower than the exact value. This is called showing the "floor" of the exact percentile (rather than
rounding off the percentile to the nearest 1 percentile). For example, if the exact percentile is equal to or
greater than 79 but less than 80, it is displayed as "79th percentile." If the exact percentile is equal to or greater
than 80 but less than 81, it is displayed as "80th percentile." The reason for this is to ensure that EJSCREEN only
displays "80th percentile" if the exact percentile truly is as high as 80.50
Ties in indicator values are fairly common, especially for the lead paint indicator and percent linguistic isolation,
where large shares are tied with values of zero. Ties are assigned a percentile that can be thought of as the
upper edge of the range of tied values. For example, if 3 percent of the US population were tied at the
maximum indicator value, they would all be shown as being at the 100th percentile, and the next lower value
would be assigned the 97th percentile. If 4 percent were all tied for the lowest value, they would all be shown
as being at the 4th percentile, and nobody would be shown as being at the 0-3 percentiles. The percent
linguistic isolation was zero in block groups comprising about 45% of the US population in the 2008-2012 ACS
data, so all those places were shown as tied for the 45th percentile, and none were reported as being at any
lower percentiles. However, it is worth noting that a group of tied values is usually not shared by more than 1%
of the population, so once the percentiles are converted to integers 0-100, the tied raw values do not cause
jumps in percentiles except in a few cases. A jump would be a case where there are no block groups assigned a
percentile between zero and 45, because around 45% of the population is tied with a value of zero.
Percentiles are assigned to calculated values (such as in buffer reports) by use of national, region-specific, and
state-specific lookup tables that show the raw value cutoff value that corresponds to each integer percentile 0-
100. To ensure that exact matches are found when looking for the 100th percentile, for example, the cutoff
values in the lookup tables are all stored with exactly 6 decimal places, and a raw value is rounded to exactly 6
decimal places before it is looked up in those tables. If a value matches the cutoff, it is assigned that percentile.
If it falls between two cutoffs, it is assigned the lower of the two percentiles, to provide displayed results that
are consistent with the way percentiles are displayed using the "floor" function described above. The lookup
tables are stored in a geodatabase used for EJCSREEN.
50 This also ensures that map colors correspond to displayed percentiles. For example, if the exact percentile is 79.99, the
map will show the place as gray, meaning it is still below the 80th percentile, and the percentile will be shown as "79th
percentile." If the map is yellow, it indicates the exact percentile is at least 80, and the displayed percentile will also be at
least 80. Without using the "floor" of the exact percentile, the map colors and displayed percentile would sometimes
disagree, and a user would not be sure if "80th percentile" actually meant the exact percentile was actually at least 80.
Using the "floor" instead of rounding ensures clarity about whether a place actually reaches a given percentile.
80 | Page
-------
Appendix B
In output tables, percentages are rounded to the nearest percent but percentiles are displayed using a "floor"
function as described above. Occasionally, this can lead to some potentially confusing situations in some
tables. For example, a place may be shown as 100% minority but only at the 98th percentile. This is because the
place may actually be 99.6% minority, which is displayed as 100% minority. But if 1.8% of the US population
lives where there is an even higher percent minority (e.g., 100%), this place is only at the 98.2 percentile, which
is displayed as 98th percentile.
The percentiles and lookup tables were calculated using the statistical software called R, using code written by
EPA, based on wtd.quantile() and wtd.Ecdf() functions in the Hmisc package (http://cran.r-
proiect.org/web/packages/Hmisc/index.html). The scripting language R is documented here: http://cran.r-
project.org
How Percentages and Raw Values are Rounded and Displayed
ESJSCREEN displays raw indicator numbers and percentiles in a standard report (on-screen or PDF format), the
"Explore reports" window, a tabular view (on-screen or downloaded text file), bar graphs, popup windows on
maps, and in the downloadable raw data files. Several standard rules have been applied to keep these formats
consistent, clear, and at an appropriate level of detail.
The raw data stored in the database used in EJSCREEN are stored as the "exact" values calculated from
estimated counts obtained from the Bureau of Census, or from the development of environmental indicators,
at the highest degree of precision used by the software calculating the indicators and by the GIS database. This
ensures that all internal calculations use the best estimate of a given number rather than relying on a rounded
off approximation, and is standard best practice in working with such data. All calculations use the "exact"
(unrounded) numbers from Census or stored in the GIS database. This includes converting raw Census data into
a demographic indicator, calculating an EJ index, or estimating the values for a buffer report.
When displaying data, such as in reports, popup windows on maps, or the tabular view, EJSCREEN presents
formatted numbers that follow certain conventions:
- Raw environmental indicator values are displayed using specified numbers of significant figures (also known
as significant digits). This is a standard way of communicating precision appropriately. Precision of these
estimates depends largely on sample size in Census survey data, and the ability of measurements and models
to estimate environmental conditions. Two significant figures are shown for all environmental indicators other
than PM2.5, ozone, and diesel PM, which are shown with three significant figures. For example, a cancer risk
calculated to be 144.44 per million would be displayed as 140 (i.e., using 2 significant figures). A PM2.5
concentration of 14.44 would be shown as 14.4 (i.e., using 3 significant figures). Proximity scores of at least
0.185 but less than 0.195 would be displayed as 0.19 (which shows 2 significant figures). This means a proximity
score displayed as 0.010 came from an exact value of at least 0.0095 but less than 0.015, for example. Note
that these significant figure rules have been applied in all cases, and if in some case the number is missing a
trailing zero that should appear, it is simply a limitation of print formatting. For example, if a proximity score is
81 | P a g e
-------
Appendix B
shown as 0.1, best practice would be to display it as 0.10 to make explicit the use of 2 significant figures, a
printout may only display it as 0.1 instead of 0.10, but it was still rounded using the 2 significant figures rule.
- Demographic percentages, such as "34% low-income," are displayed as rounded to the nearest 1%. For
example, any values equal to or greater than 79.5% but less than 80.5% are displayed as "80%."
It is also important to keep in mind that all of the numbers are estimates, so small differences in raw values or
percentiles should not be regarded as certain and meaningful, given the uncertainty in the environmental and
demographic estimates.
Calculations for Buffer Reports
EJSCREEN allows a user to define a buffer, such as the circle that includes everything within 1 mile of a specific
point. Non-circular, user-defined shapes also can be defined to represent buffers of any shape. A report
summarizes the demographics of residents within this buffer, as well as the environmental indicators and EJ
index values within the buffer.
The summary within a buffer is designed to represent the average resident within the buffer, and also provides
an estimate of the total population residing in the buffer. For example, the traffic proximity indicator for a
buffer is the population-weighted average of all the traffic indicator values in the buffer. Similarly, the percent
minority would be a weighted average, which is the same as the overall percent minority for all residents in the
buffer.
Some block groups will be partly inside and partly outside a buffer, and any buffer analysis must estimate how
much of each block group's population is inside the buffer. Areal apportionment of block groups is one
standard method, but it assumes that population is evenly spread throughout a block group, which may be far
from the actual distribution of residents. Areal apportionment of blocks would be even more accurate but
extremely computationally intensive.
To provide the most accurate counts that are currently feasible for a screening tool, EJSCREEN uses an
approach based on Census block internal points. EJSCREEN estimates the fraction of the Census block group
population that is inside the buffer by using block-level population counts from Census 2010. These blocks
provide data about where residents are at a higher resolution than block groups. Each block has an internal
point defined by the Census Bureau, and the entire block population is counted as inside or outside the buffer
depending on whether the block internal point is inside or outside. This assumption typically introduces
relatively little error because blocks are so small relative to a typical buffer, so a small fraction of the total
buffer population is in blocks that span an edge of the buffer. Also, any blocks along the edge of a buffer whose
populations are close to 0 or 100% inside the buffer will be well represented by this assumption.
As long as users draw buffers much larger than a local block group, this method should represent the average
person inside the buffer reasonably well.
82 | P a g e
-------
Appendix B
The calculation of a value for the buffer is essentially the population-weighted average of the indicator values
in the blocks included in the buffer, where each block uses the indicator values of the block group containing it,
A block group is weighted based on the fraction of the ACS block group population that is considered in the
buffer. That fraction is estimated as the Census 2010 block population divided by the Census 2010 block group
population. The formula below is used to estimate the population average of a raw indicator value in a buffer.
This formula is simply a population-weighted average - it sums the population-weighted raw values, and then
divides that sum by the total population in the buffer.
"BlockPoplO" refers to the Census 2010 block level population total (used here because the ACS does not
provide block resolution), and "BG" indicates block group. "BGACSPop" is the block group estimated population
count from the ACS, which is often different than the Census 2010 total for all blocks in the block group,
because the ACS data used here is a composite estimate based on survey samples spanning five years, while
the Census is a full count at one point in time.
Demographic Data and Geographic Coverage
In the first decade of this century, the Census Bureau made a fundamental shift in how detailed demographic
data are collected. Rather than collecting basic data from everyone, plus more detailed data from a one-in-six
sample of households once a decade in the decennial census, a mixed approach has been adopted. The basic
data, required for Congressional redistricting under the U.S. Constitution, are still collected every ten years in
what is intended to be a 100% census. But that basic information, plus virtually all the more detailed
demographic data, are also collected throughout each year in a stratified random sample of more than 200,000
households each month. This is the American Community Survey (ACS). Some of this information is then
aggregated and displayed in yearly summaries, others in 3-year summaries, and others in 5-year summaries.
Only the five-year summary files provide block group resolution. The result is a timelier, evolving picture of U.S.
demographics. For instance, the ACS 2005 to 2009 average data were released in December 2010, while most
demographics data users were still working with the April 2000 decennial census snapshot.
Extensive documentation of the ACS is available. For a general overview, see
http://www.census.gov/acs/www/ and for complete documentation see
http://www.census.gov/acs/www/data documentation/documentation main/. For information on the 5-year
summary file, which is what EJSCREEN uses, see
http://www.census.gov/acs/www/data documentation/summary file/. For information on using the data see
http://www.census.gov/acs/www/guidance for data users/handbooks/).
Race and ethnicity, the two items that determine minority status in our approach, are available from the 100%
enumeration from the decennial census or the ACS, while all the other measures are only contained in ACS
estimates. For the purposes of EJSCREEN, EPA did not believe that the increased precision of the minority
Faie(A)= £ _BCPovW
BlockPoplO
* BGACSPop * BG_RawValue
VBlktflknA 2-iVBlk£lkr\A
V
BlkPoplQ
BGPoplG *BGACSP°P
83 | P a g e
-------
Appendix B
measures that might be gained from combining those once per decade data with the other data items from
ACS were worth the problems and ambiguities entailed in such a hybrid approach.
All of EJSCREEN's demographic data come from the latest annual update of the five-year average ACS
estimates, with some lag time from publication by Census to inclusion in EJSCREEN. The Census Bureau does
not recommend making direct comparisons between data from a five-year summary and a prior, overlapping
five-year period, such as comparing 2007-2011 to 2008-2012.51 This means attempting to look for trends in
terms of year-to-year changes is not recommended - changes in ACS estimates at block group resolution can
be reviewed every five years, but not more often. Yearly changes can be examined at county resolution using
the 1-year ACS data.
Each of the nation's counties (or county equivalents, such as Municipios in Puerto Rico) is completely divided
into Census tracts. Each tract is in turn divided into Census block groups. Census block groups generally have
between 600 and 3,000 people, with an optimal population of 1,500; however, a few are much more populous
and a small number have zero residents. A block group consists of one or more Census blocks. In urban areas, a
block is typically a city block defined by streets. The Census Bureau collects data by household, but block
groups are the smallest area for which the ACS presents estimates.
Tracts and block groups are defined in ACS exactly as in the Census 2010 other than a few exceptions due to
updates or corrections.52 For the 2008-2012 ACS data in EJSCREEN, Census made changes in a few dozen block
groups and their tracts and FIPS codes in NY, AZ, and CA53. Also see details in the discussion of how NATA data
was converted to 2010 boundaries, in Section 3.
There are 11,078,297 blocks in the Census 2010 data, not including Puerto Rico and the Island Areas,54 but
since then some changes have been made in the FIPS codes that relate blocks to their parent block groups.
Block to block group relationships are used by EJSCREEN in part of the buffer analysis (as described in Appendix
C). This required making manual adjustments to reassign some Census 2010 blocks to their updated parent ACS
block groups for purposes of calculating buffer estimates.
The EJSCREEN dataset based on the ACS 2008-2012 summary file has data for 51 States/equivalents (includes
DC), 3,143 counties/county equivalents, 73,056 census tracts, and 217,739 block groups. Puerto Rico (2,594
block groups) is part of the ACS, but could not be included in the 2015 version of the EJSCREEN dataset due to
environmental data and other constraints, although it may be included in future versions. The Census Bureau
does not collect ACS data for the Virgin Islands or the other Island Territories.
Compared to the ACS 2008-2012, the Census 2010 tallies counted one block group more than are included in
the ACS (or EJSCREEN) - an extra block group in New York State - but they were otherwise identical in block
51 http://www.census.gov/acs/www/guidance for data users/comparing data/
52 http://www.census.gov/acs/www/data documentation/geography/
53 See https://www.census.gov/acs/www/data documentation/2011 geography release notes/ on NY changes and
https://www.census.gov/acs/www/data documentation/geography notes/index.php on AZ and CA changes.
54 http://www.census.gov/geo/maps-data/data/tallies/tractblock.html
84 | Page
-------
Appendix B
group counts per state. The count of geographies (e.g., number of block groups) covered by the ACS is
summarized by the Census Bureau for each update:
ACS tallies, as counted in data downloaded from ACS FTP site, and as used in EJSCREEN (217,739 block
groups, without Puerto Rico)
ACS tallies, with Puerto Rico55: 220,333 with PR, 217,739 without Puerto Rico.
Census 2010 tallies, by State/PR56: 220,334 with Puerto Rico, and 217,740 without Puerto Rico.
Table 4. Tallies of 2008-12 ACS Block Groups Used in 2015 Version of EJSCREEN
Geography
Number of Block
Groups
In EJSCREEN?
Continental U.S., 48 States plus DC
216,330
Yes
Alaska & Hawaii
534 & 875
Yes
SUBTOTAL: Included in EJSCREEN
217,739
Yes
Puerto Rico
2,594
No
Virgin Islands and other Island Territories
(American Samoa, Commonwealth of the
Northern Mariana Islands, Guam)
408
No
Source: U.S. Census Bureau
http://www.census.gov/acs/www/data documentation/areas published/ and
http://www.census.gov/geo/maps-data/data/tallies/tractblock.html
Demographic Variables and Formulas
This section provides details on the Census variables and formulas used to calculate demographic indicators for
each block group. Short variable names used in EJSCREEN internal calculations are shown below, preceded by
tabular data showing the ACS summary file table number, sequence number, variable number, and name of the
table or variable. For example, total population, referred to in EJSCREEN calculations as "pop" and called
"ACSTOTPOP" in the geodatabase, is taken from the Census variable B01001.001, in ACS 5-year summary file
Table B01001. Field names as used in the EJSCREEN geodatabase differ from the names shown below, and a
table was used to map between alternative fieldnames.
These details are based on the ACS 2008-2012 summary files.
55 http://www.census.gov/acs/www/data documentation/areas published/
56 http://www.census.gov/geo/maps-data/data/tallies/tractblock.html
85 | P a g e
-------
ACS Summary File documentation is here:
http://www.census.gov/acs/www/data documentation/summary
-------
Appendix B
ACS Summary File Variables
Tables used from ACS 2008-2012 are shown below. Block group data were obtained from the Census FTP site57,
and are not available from American Fact Finder at block group resolution.
Note that only selected variables from these tables are in the EJSCREEN geodatabase, for performance reasons.
Many of the intermediate variables or detailed breakdowns are not in the geodatabase. Documentation
supplied with the geodatabase download explains the variable names used in the geodatabase.
The URLs that can be used to view and download one ACS 2008-2012 table at a time, for the US total only, are
as follows:
Table 5. ACS Tables Underlying EJSCREEN Demographic Data and Lead Paint Indicator
ACS
Table
ID
URL for US summary table
via American Fact Finder
Table Title
B01001
http://factfinder.census.Rov/bkmk/table
SEX BY AGE
/1.0/en/ACS/12 5YR/B01001/0100000US
B03002
http: / /factfinder, census.eov/bkmk/table
HISPANIC OR LATINO ORIGIN BY RACE
/1.0/en/ACS/12 5YR/B03002/0100000US
B15002
http: / /factfinder, census. eov/bkmk/table
SEX BY EDUCATIONAL ATTAINMENT FOR THE
POPULATION 25 YEARS AND OVER
/1.0/en/ACS/12 5YR/B15002/0100000US
B16002
http: //factfinder, census .eov/bkmk/table
HOUSEHOLD LANGUAGE BY HOUSEHOLDS IN WHICH NO
ONE 14 AND OVER SPEAKS ENGLISH ONLY OR SPEAKS
A LANGUAGE OTHER THAN ENGLISH AT HOME AND
SPEAKS ENGLISH "VERY WELL"
/1.0/en/ACS/12 5YR/B16002/0100000US
C17002
http: //factfinder, census .eov/bkmk/table
RATIO OF INCOME TO POVERTY LEVEL IN THE PAST
12 MONTHS
/I.0/en/ACS/12 5YR/C17002/0100000US
B25034
http: //factfinder, census .eov/bkmk/table
YEAR STRUCTURE BUILT
/1.0/en/ACS/12 5YR/B25034/0100000US
URLs to view and download one ACS 2008-2012 table at a time, for the totals for the US,
every state plus DC, and Puerto Rico are shown below.
57 http://www.census.gov/acs/www/data documentation/data via ftp/
87 | P a g e
-------
Appendix B
B01001
http://factfinder2.census.gov/bkmk/table/1.0/en/ACS/12 5YR/B01001/0100000US10400000US0110400000US0210400000US0410400000US051040000
0US0610400000US08 10400000US09 10400000US1010400000US1110400000US1210400000US13 10400000US15 10400000US1610400000US1710400000U
S18I0400000US19I0400000US20I0400000US21I0400000US22I0400000US23I0400000US24I0400000US25I0400000US26I0400000US27I0400000US2
8 10400000US2910400000US3010400000US3110400000US3210400000US3310400000US3410400000US3510400000US3610400000US3710400000US38I
0400000US3910400000US4010400000US4110400000US42 10400000US4410400000US4510400000US4610400000US4710400000US48 10400000US49 104
00000US5010400000US5110400000US5310400000US5410400000US5510400000US5610400000US72
B03002
http://factfinder2.census.gOv/bkmk/table/l.0/en/ACS/12 5YR/B03002/0100000US10400000US0110400000US0210400000US0410400000US051040000
0US0610400000US08 10400000US09 10400000US1010400000US1110400000US1210400000US13 10400000US15 10400000US1610400000US1710400000U
S18I0400000US19I0400000US20I0400000US21I0400000US22I0400000US23I0400000US24I0400000US25I0400000US26I0400000US27I0400000US2
8 10400000US2910400000US3010400000US3110400000US3210400000US3310400000US3410400000US3510400000US3610400000US3710400000US38I
0400000US3910400000US4010400000US4110400000US42 10400000US4410400000US4510400000US4610400000US4710400000US48 10400000US49 104
00000US5010400000US5110400000US5310400000US5410400000US5510400000US5610400000US72
B15002
http://factfinder2.census.gOv/bkmk/table/l.0/en/ACS/12 5YR/B15002/0100000US10400000US0110400000US0210400000US0410400000US051040000
0US0610400000US08 10400000US09 10400000US1010400000US1110400000US1210400000US13 10400000US15 10400000US1610400000US1710400000U
S18I0400000US19I0400000US20I0400000US21I0400000US22I0400000US23I0400000US24I0400000US25I0400000US26I0400000US27I0400000US2
8 10400000US29 10400000US3010400000US3110400000US3210400000US33 10400000US3410400000US35 10400000US3610400000US3710400000US38 I
0400000US3910400000US4010400000US4110400000US42 10400000US4410400000US45 10400000US4610400000US4710400000US48 10400000US49 104
00000US5010400000US5110400000US5310400000US5410400000US5510400000US5610400000US72
B16002
http://factfinder2.census.gov/bkmk/table/1.0/en/ACS/12 5YR/B16002/0100000US10400000US0110400000US0210400000US0410400000US051040000
0US0610400000US08 10400000US09 10400000US1010400000US1110400000US1210400000US13 10400000US15 10400000US1610400000US1710400000U
S18 10400000US19 10400000US2010400000US2110400000US22 10400000US23 10400000US2410400000US2510400000US2610400000US2710400000US2
8 10400000US2910400000US3010400000US3110400000US3210400000US3310400000US3410400000US3510400000US3610400000US3710400000US38I
0400000US3910400000US4010400000US4110400000US42 10400000US4410400000US4510400000US4610400000US4710400000US48 10400000US49 104
00000US5010400000US5110400000US5310400000US5410400000US5510400000US5610400000US72
C17002
http://factfinder2.census.gov/bkmk/table/1.0/en/ACS/12 5YR/C17002/0100000US10400000US0110400000US0210400000US0410400000US051040000
0US0610400000US08 10400000US09 10400000US1010400000US1110400000US1210400000US13 10400000US15 10400000US1610400000US1710400000U
S18I0400000US19I0400000US20I0400000US21I0400000US22I0400000US23I0400000US24I0400000US25I0400000US26I0400000US27I0400000US2
8 10400000US29 10400000US3010400000US3110400000US32 10400000US33 10400000US3410400000US35 10400000US3610400000US3710400000US38 I
0400000US3910400000US4010400000US4110400000US42 10400000US4410400000US45 10400000US4610400000US4710400000US48 10400000US49 104
00000US5010400000US5110400000US5310400000US5410400000US5510400000US5610400000US72
B25034
http://factfinder2.census.gOv/bkmk/table/l.0/en/ACS/12 5YR/B25034/0100000US10400000US0110400000US0210400000US0410400000US051040000
0US0610400000US08 10400000US09 10400000US1010400000US1110400000US1210400000US13 10400000US15 10400000US1610400000US1710400000U
S18I0400000US19I0400000US20I0400000US21I0400000US22I0400000US23I0400000US24I0400000US25I0400000US26I0400000US27I0400000US2
8 10400000US2910400000US3010400000US3110400000US3210400000US3310400000US3410400000US3510400000US3610400000US3710400000US38I
0400000US39 10400000US4010400000US4110400000US42 10400000US4410400000US45 10400000US4610400000US4710400000US48 10400000US49 104
00000US5010400000US5110400000US5310400000US5410400000US5510400000US5610400000US72
88 | P a g e
-------
Appendix B
TOTAL POPULATION COUNTS AND AGES
SEX BY AGE
Table.ID Sequence.Number Line.Number Field
B01001
0002
NA
B01001
0002
NA Universe:
Total population
B01001
0002
1
Total:
B01001
0002
2
Male:
B01001
0002
3
Under 5 years
B01001
0002
4
5 to 9 years
B01001
0002
5
10 to 14 years
B01001
0002
6
15 to 17 years
B01001
0002
7
18 and 19 years
B01001
0002
8
20 years
B01001
0002
9
21 years
B01001
0002
10
22 to 24 years
B01001
0002
11
25 to 29 years
B01001
0002
12
30 to 34 years
B01001
0002
13
35 to 39 years
B01001
0002
14
40 to 44 years
B01001
0002
15
45 to 49 years
B01001
0002
16
50 to 54 years
B01001
0002
17
55 to 59 years
B01001
0002
18
60 and 61 years
B01001
0002
19
62 to 64 years
B01001
0002
20
65 and 66 years
B01001
0002
21
67 to 69 years
B01001
0002
22
70 to 74 years
B01001
0002
23
75 to 79 years
B01001
0002
24
80 to 84 years
B01001
0002
25
85 years and over
B01001
0002
26
Female:
B01001
0002
27
Under 5 years
B01001
0002
28
5 to 9 years
B01001
0002
29
10 to 14 years
B01001
0002
30
15 to 17 years
B01001
0002
31
18 and 19 years
B01001
0002
32
20 years
B01001
0002
33
21 years
B01001
0002
34
22 to 24 years
B01001
0002
35
25 to 29 years
B01001
0002
36
30 to 34 years
B01001
0002
37
35 to 39 years
B01001
0002
38
40 to 44 years
89 | P a g e
-------
Appendix B
B01001
0002
39
45 to 49 years
B01001
0002
40
50 to 54 years
B01001
0002
41
55 to 59 years
B01001
0002
42
60 and 61 years
B01001
0002
43
62 to 64 years
B01001
0002
44
65 and 66 years
B01001
0002
45
67 to 69 years
B01001
0002
46
70 to 74 years
B01001
0002
47
75 to 79 years
B01001
0002
48
80 to 84 years
B01001
0002
49
85 years and over
pop= B01001.001
ageunder5m = B01001.003
age5to9m = B01001.004
agel0tol4m = B01001.005
agel5tol7m = B01001.006
age65to66m = B01001.020
age6769m = B01001.021
age7074m = B01001.022
age7579m = B01001.023
age8084m = B01001.024
age85upm = B01001.025
ageunder5f = B01001.027
age5to9f = B01001.028
agel0tol4f = B01001.029
agel5tol7f= B01001.030
age65to66f = B01001.044
age6769f = B01001.045
age7074f = B01001.046
age7579f = B01001.047
age8084f = B01001.048
age85upf = B01001.049
90 | P a g e
-------
Appendix B
RACE/ETHNICITY
HISPANIC OR LATINO ORIGIN BY RACE
Table.ID Sequence.Number Line.Number Field
B03002
0005
NA
B03002
0005
NA
Universe: Total population
B03002
0005
1
Total:
B03002
0005
2
Not Hispanic or Latino:
B03002
0005
3
White alone
B03002
0005
4
Black or African American alone
B03002
0005
5
American Indian and Alaska Native alone
B03002
0005
6
Asian alone
B03002
0005
7
Native Hawaiian and Other Pacific Islander alone
B03002
0005
8
Some other race alone
B03002
0005
9
Two or more races:
B03002
0005
10
Two races including Some other race
B03002
0005
11
Two races excluding Some other race, and three or
races
B03002
0005
12
Hispanic or Latino:
B03002
0005
13
White alone
B03002
0005
14
Black or African American alone
B03002
0005
15
American Indian and Alaska Native alone
B03002
0005
16
Asian alone
B03002
0005
17
Native Hawaiian and Other Pacific Islander alone
B03002
0005
18
Some other race alone
B03002
0005
19
Two or more races:
B03002
0005
20
Two races including Some other race
B03002
0005
21
Two races excluding Some other race, and three or
races
pop3002 = B03002.001
nhwa = B03002.003
EDUCATIONAL ATTAINMENT FOR THOSE AGE 25+
SEX BY EDUCATIONAL ATTAINMENT FOR THE POPULATION 25 YEARS AND OVER
Table.ID Sequence.Number Line.Number Field
B15002 0043 NA
B15002 0043 NA Universe: Population 25 years and over
B15002 0043 1 Total:
91 | Page
-------
Appendix B
B15002
0043
2
Male:
B15002
0043
3
No schooling completed
B15002
0043
4
Nursery to 4th grade
B15002
0043
5
5th and 6th grade
B15002
0043
6
7th and 8th grade
B15002
0043
7
9th grade
B15002
0043
8
10th grade
B15002
0043
9
11th grade
B15002
0043
10
12th grade, no diploma
B15002
0043
11
High school graduate, GED, or alternative
B15002
0043
12
Some college, less than 1 year
B15002
0043
13
Some college, 1 or more years, no degree
B15002
0043
14
Associate's degree
B15002
0043
15
Bachelor's degree
B15002
0043
16
Master's degree
B15002
0043
17
Professional school degree
B15002
0043
18
Doctorate degree
B15002
0043
19
Female:
B15002
0043
20
No schooling completed
B15002
0043
21
Nursery to 4th grade
B15002
0043
22
5th and 6th grade
B15002
0043
23
7th and 8th grade
B15002
0043
24
9th grade
B15002
0043
25
10th grade
B15002
0043
26
11th grade
B15002
0043
27
12th grade, no diploma
B15002
0043
28
High school graduate, GED, or alternative
B15002
0043
29
Some college, less than 1 year
B15002
0043
30
Some college, 1 or more years, no degree
B15002
0043
31
Associate's degree
B15002
0043
32
Bachelor's degree
B15002
0043
33
Master's degree
B15002
0043
34
Professional school degree
B15002
0043
35
Doctorate degree
age25up = B15002.001
mO = B15002.003 (males age 25+ with zero education)
m4 = B15002.004 (males age 25+ with >0 up to 4th grade)
m6 = B15002.005
m8 = B15002.006
m9 = B15002.007
mlO = B15002.008
mil = B15002.009
ml2 = B15002.010 (males age 25+ with high school diploma)
92 | P a g e
-------
Appendix B
fO = B15002.020
f4 = B15002.021
f6 = B15002.022
f8 = B15002.023
f9 = B15002.024
flO= B15002.025
fll= B15002.026
fl2 = B15002.027
HOUSEHOLDS THAT ARE LINGUISTICALLY ISOLATED
"HOUSEHOLD LANGUAGE BY HOUSEHOLDS IN WHICH NO ONE 14 AND OVER SPEAKS ENGLISH ONLY OR
SPEAKS A LANGUAGE OTHER THAN ENGLISH AT HOME AND SPEAKS ENGLISH "VERY WELL"
Table.ID Seq.Number Line.Number Field
B16002 0044 NA
B16002 0044 NA "Universe: Households"
B16002 0044 1 "Total:"
B16002
0044
2
"English only"
B16002
0044
3
"Spanish:"
B16002
0044
4
No one 14 and over speaks English only or speaks English "very well"
B16002 0044 5
At least one person 14 and over speaks English only or speaks English "very well"
B16002 0044 6 Other Indo-European languages:
B16002 0044 7
No one 14 and over speaks English only or speaks English "very well"
B16002 0044 8
At least one person 14 and over speaks English only or speaks English "very well"
B16002 0044 9 Asian and Pacific Island languages:
B16002 0044 10
No one 14 and over speaks English only or speaks English "very well"
B16002 0044 11
At least one person 14 and over speaks English only or speaks English "very well"
B16002 0044 12 Other languages:
B16002 0044 13
No one 14 and over speaks English only or speaks English "very well"
B16002 0044 14
At least one person 14 and over speaks English only or speaks English "very well"
93 | P a g e
-------
Appendix B
hhlds = B16002.001
lingisospanish = B16002.004
lingisoeuro = B16002.007
lingisoasian = B16002.010
lingisoother = B16002.013
INDIVIDUALS BY RATIO OF INCOME TO POVERTY THRESHOLD
RATIO OF INCOME TO POVERTY LEVEL IN THE PAST 12 MONTHS
Table.ID
Sequence.Number
Line
.Number
Field
C17002
0049
NA
C17002
0049
NA
Universe:
Population for whom poverty
status is determined
C17002
0049
1
Total:
C17002
0049
2
Under .50
C17002
0049
3
.50 to .99
C17002
0049
4
1.00 to 1.24
C17002
0049
5
1.25 to 1.49
C17002
0049
6
1.50 to 1.84
C17002
0049
7
1.85 to 1.99
C17002
0049
8
2.00 and over
povknownratio = C17002.001
pov50 = C17002.002 (below 0.50 times poverty threshold)
pov99 = C17002.003 (0.5 to 0.99 times poverty threshold)
povl24 = C17002.004
povl49 = C17002.005
povl84 = C17002.006
povl99 = C17002.007
pov2plus = C17002.008
AGE OF OCCUPIED HOUSING UNITS (CORRELATED WITH LEAD PAINT)
YEAR STRUCTURE BUILT
Table.ID Sequence.Number Line.Number Field
B25034 0104 NA
94 | P a g e
-------
Appendix B
B25034
B25034
B25034
B25034
B25034
B25034
B25034
B25034
B25034
B25034
B25034
0104
0104
0104
0104
0104
0104
0104
0104
0104
0104
0104
NA
1
2
3
4
5
6
7
8
9
10
Universe: Housing units
Total:
Built 2010 or later
Built 2000 to 2009
Built 1990 to 1999
Built 1980 to 1989
Built 1970 to 1979
Built 1960 to 1969
Built 1950 to 1959
Built 1940 to 1949
Built 1939 or earlier
builtunits = B25034.001
builtl950tol959 = B25034.008
builtl940tol949 = B25034.009
builtprel940 = B25034.010
Calculated Demographic Data Fields
Based on the raw counts from the ACS described above, various demographic variables were calculated for use
in EJSCREEN. Conditional formulas below are in R syntax, and generally indicate that a value of zero was used in
cases where the denominator was zero, to avoid division by zero. For example, the formula "pctmin =
ifelse(pop==0,0, as.numeric(mins ) / pop)" indicates that percent minority was calculated as the ratio of
number of minorities over total population of a block group, but was set to zero if the population was zero.
# RACE/ETHNICITY COMBINED, CALCULATED VARIABLES
mins = pop - nhwa
pctmin = ifelse(pop==0,0, as.numeric(mins ) / pop)
# POVERTY, LOW-INCOME CALCULATED VARIABLES
# poverty ratios
num2pov = numlpov + povl24 + povl49 + povl84 + povl99
lowinc = num2pov
pct2pov = ifelse( povknownratio==0,0, num2pov/povknownratio)
pctlowinc = pct2pov
num2pov.alt = povknownratio - pov2plus
pct2pov.alt = ifelse( povknownratio==0,0, num2pov.alt/povknownratio)
95 | P a g e
-------
Appendix B
# EDUCATIONAL ATTAINMENT CALCULATED VARIABLES
Iths = mO + m4 + m6 + m8 + m9 + mlO + mil + ml2 +
fO + f4 + f6 + f8 + f9 + flO + fll + fl2
pctlths = ifelse(age25up==0,0, as.numeric(lths ) / age25up)
# LINGUISTIC ISOLATION CALCULATED VARIABLES
lingiso = lingisospanish + lingisoeuro + lingisoasian + lingisoother
pctlingiso = ifelse( hhlds==0,0, lingiso / hhlds)
# AGE GROUPS CALCULATED VARIABLES
under5 = ageunder5m + ageunder5f
pctunder5 = ifelse( pop==0,0, under5/pop)
over64 = age65to66m + age6769m + age7074m + age7579m + age8084m + age85upm +
age65to66f + age6769f + age7074f + age7579f + age8084f + age85upf
pctover64 = ifelse( pop==0,0, over64/pop)
# HOUSING CALCULATED VARIABLES (LEAD PAINT INDICATOR)
prel960 = builtprel940 + builtl940tol949 + builtl950tol959
pctprel960 = ifelse( builtunits==0,0, prel960/builtunits)
96 | P a g e
-------
Appendix B
Uncertainty and Limitations in the Demographic Data
Uncertainty in demographic data
As with every sample survey, sampling results in unavoidable approximations in every estimate that comes
from the survey. The Census Bureau clearly labels every data item as an "estimate," and accompanies each
with an estimate of its margin of error. Anyone using a screening tool should be aware of those demographic
uncertainties, together with uncertainties in the environmental measures, in tables, graphical displays and
descriptive materials.
Uncertainties are also discussed in section 1 (as general caveats), section 2 (with regard to buffer reports), and
Appendix B (in discussions of buffering details and demographic data).
Users of EJSCREEN must keep in mind the substantial uncertainty in estimated demographic and environmental
indicators used in screening tools such as EJSCREEN. Uncertainty is a critical consideration when using
EJSCREEN because the tool relies on demographic and environmental estimates at block group resolution. As
the Census Bureau makes clear in documentation of the American Community Survey (ACS), the margin of
error for an estimate in a given block group is often very large relative to the estimate, so an estimate of
percent low-income, for example, is often very uncertain for a single block group.
Combined with uncertainty in environmental data, this means EJ index values are often very uncertain at block
group resolution. Therefore, modest differences in percentile scores between block groups or small buffers
should not be interpreted as meaningful because of the uncertainties in demographic and environmental data
at the block group level. We do not have a high degree of confidence when comparing or ranking places with
only modest differences in estimated percentile. For this reason, it is critical that EJSCREEN results be
interpreted carefully and that additional information be used to supplement or follow up on screening, where
appropriate. Section 1 of this document discusses caveats and limitations further.
EPA cannot provide precise confidence intervals on EJ indexes or percentiles due to technical limitations in the
data made public by the Census Bureau and the challenges of quantifying uncertainty for the environmental
indicators. Technical documentation on methods and challenges in estimating uncertainty for calculated
demographic indicators using the ACS is available from the Census Bureau58 (with challenges described in
related technical documents59). ESRI also provides useful discussions of margin of error.60
It is likely that block group errors in the various data fields reported by Census (e.g., count with income-poverty
ratio below 0.5, count with ratio 0.5 to 1, etc.) are correlated. Relevant covariances, however, are not provided
by the Census Bureau. This means simple methods of approximating margin of error for a calculated variable
(e.g., percent low-income) may not be entirely adequate. In this case, it appears that a custom tabulation by
58http://www.census.gov/acs/www/Downloads/data documentation/Statistical Testing/2011StatisticalTesting3and5vear
.pdf
59 http://www.census.gov/acs/www/Downloads/data documentation/Accuracv/MultivearACSAccuracvofData2011.pdf
60 http://www.esri.com/software/american-communitv-survev/understanding-margin-error
97 | P a g e
-------
Appendix B
the Census Bureau would be the most accurate way to generate reliable estimates of the margin of error for
variables such as the percent low-income or the demographic indicator, for use in creating confidence intervals
around an EJ index or the percentile of that index. Future research may be able to produce reasonable
approximations of confidence intervals around block group or buffer estimates. EJSCREEN users should keep in
mind that using a buffer larger than the local block groups will produce more reliable estimates than a single
block group can provide.
Using 2x poverty rate
The rationale for using twice the poverty threshold rather than just the poverty threshold includes the
following considerations:
The effects of income on baseline health and probably on other aspects of susceptibility are not
limited to those below the poverty thresholds those from lx to 2x poverty also have worse
health overall than those with higher incomes (Centers for Disease Control and Prevention, 2010),
and asthma rates, for example, begin to increase as income falls below twice the poverty threshold
(Centers for Disease Control and Prevention, 2011a).
Many studies in various fields use 2x poverty, and many others use lx poverty (e.g., see Su et al.,
2009); the same is true for prior EPA screening tools. There is precedent for both. However, a
rationale often mentioned is that today's poverty thresholds are too low to adequately capture the
populations adversely affected by low income levels, especially in high-cost areas. Some analysts
have concluded that the amount of income actually required for basic living costs without
government support is far higher than the current Federal poverty thresholds (Cauthen & Fass,
2008).
When using twice the poverty threshold, the number or percent low income happens to roughly
equal number or percent minority in the United States. This makes it convenient and simple to use
the average of the two without applying any other weights to them, and in this way each low-
income person affects the susceptibility indicator about as much as each minority person.
The Census Bureau has been developing experimental poverty measures that account for local
costs of living, but these are not yet in widespread use.61
Interpretation of Demographic Indexes
The demographic indexes are meant to reflect some of the combined impacts of multiple demographic factors.
The Census Bureau does not provide a tabulation of low-income residents by race/ethnicity at the block group
level62, so it is impossible to know what percentage of a block group is low-income minorities vs. low-income
non-minorities, for example. EJSCREEN simply defines the demographic index as the average of the percentage
of people who are low income and the percentage of people who are minorities. Therefore, this demographic
61 https://www.census.gov/hhes/povmeas/
62 Table B17001 and related tables provide tract-level cross-tabulations of race-ethnicity and poverty, but not percent low-
income as defined in EJSCREEN.
98 | Page
-------
Appendix B
index will be (equal to or) smaller than the percentage of people who are in at least one of these groups. In
other words, it is typically smaller than the share of people who are in one or more of these groups - just low
income, just minority, or both. The average will also be (equal to or) larger than the percentage of people who
are simultaneously in all of these groups. It is larger than (or equal to) the share of people who are
simultaneously low income and minority. The value of the demographic index is almost always larger than the
number of people who are simultaneously minority and low-income, because usually some people are in only
one of these demographic groups. Note that one person cannot be under five and over 64, so any one person
can be in up to five of the six demographic groups used in EJSCREEN.
The demographic index is also bounded by these two percentages (percent low income and percent minority).
For example, the actual percentage minority is larger than the value of the demographic index, if the
percentage low-income is lower than the percentage minority.
99 | P a g e
-------
Appendix C
APPENDIX C. TECHNICAL DETAILS ON PROXIMITY INDICATORS
Several of EJSCREEN's environmental factors are direct or indirect estimates of potential exposure or health
risks, such as the NATA cancer risk estimates and the ozone and PM2.5 concentration estimates. There are other
aspects of an individual's or a community's environmental concerns that are less readily quantified in terms of
emissions, concentrations, or risk estimates.
People may be concerned about living near facilities that handle hazardous substances, and other potential
sources of pollution, such as highways or abandoned waste sites. Concern over "locally undesirable land uses",
or LULUs, is in some cases founded on the potential for routine or episodic releases of pollutants to the air, land
or water, and the potential for such releases to cause human health or environmental adverse effects or other
societal disamenities.
The purpose of the proximity measures in EJSCREEN is to systematically and consistently quantify different
degrees of potential for these effects. We have developed a method to calculate a score that represents the
relative magnitude of the proximity of the population within a block group to facilities, waste sites, or traffic
surrounding it. A block group with more facilities closer to the block group's residential population will have a
higher score than a block group where facilities are further away. We have applied this method to these facility
or site types:
National Priorities List (NPL) sites (a key subset of "Superfund," sites).
Hazardous waste Treatment, Storage or Disposal Facilities (TSDFs), subject to regulations under the
Resource Conservation and Recovery Act (RCRA).
Risk Management Plan (RMP) facilities, which are facilities that maintain greater than certain
quantities of extremely hazardous substances, and are required to take certain actions, including
filing risk management plans, under section 112 (r) of the Clean Air Act.
Major direct dischargers to water permitted under the National Pollutant Discharge Elimination
System (NPDES).
We have developed a similar approach to represent proximity to and traffic volume on nearby highways.
In the sections below, we will describe the general approach, in terms of facility proximity. We will then
describe how it differs for traffic proximity. Then we will discuss certain adjustments we have made, mostly to
make the approach computationally efficient, and summarize the data sources and computational routine that
we applied to implement this approach. We conclude with caveats and other observations.
Calculating Proximity to Facilities or NPL Sites
Each of the more than 217,000 block groups for the U.S. states and the District of Columbia is made up of
between one block and several hundred blocks. Most block groups nationwide are smaller than approximately
0.5 square miles, an area that if circular would have a radius of about 640 meters. In block groups of this
median size, the average residence generally would be about 430 to 720 meters (or less than half a mile) away
100 | P a g e
-------
Appendix C
from a given point within the block group, such as a facility, as explained at the end of Appendix C. About 20-
25% of block groups covered an area smaller than a circle of radius 300-350 meters (almost one quarter of a
mile), as of the 2005-2009 geographies. Also, a very small number of block groups are extremely large in area,
in very rural locations.
All of a block group's blocks may have residential population estimated by the 5-year ACS, or only some, and
some block groups have no residents at all. Blocks and block groups vary greatly in geographic area, and in
population. The approach used here works first at the block level, based on measures of proximity to the
facilities in or near the blocks. The block-level measures are then aggregated among all the blocks within a
block group, weighted by the number of people in the different blocks.
Thus, while population is considered in aggregating the block scores, the measure does not increase or
decrease for block groups with higher or lower populations. The measure is, rather, a characteristic of the
residents of the block group, in the same way that cancer risk from NATA or ozone concentration are estimated
measures of the conditions of those places.
Let
i represent a particular facility
j represent a block within a block group
k represent a block group
dij is the distance, in kilometers, from block j's centroid to the given location of facility i
popjk is the estimated population of block j within block group k
popk is the total estimated population of block group k
f(dij) is a function representing the proximity of facilty i to block j, a declining function of the distance,
dij
BlockScorejk is the aggregation of the proximity influences of all facilities affecting block jk
BlockGroupScorek is the population-weighted aggregation of the block group's component blocks
We have chosen to define the proximity function as
f(dij) = 1 / dij
That is, a facility 1 kilometer from a block's population contributes twice the score as a facility 2 kilometers
from the same block. We note that we have made a choice in using inverse distance for this function. Air
dispersion modeling for pollutants following Gaussian plume assumptions would show a generally greater
drop-off in concentration, roughly with the second power to 2.5 power of one over distance. But actual
concentrations around individual plants follow often-complex patterns that depend on the particular mix of
stack vs. fugitive emissions, characteristics of stack height, exit velocity and temperature, the presence of
buildings or other land surface characteristics and meteorology. Some substances react readily with other
substances in the atmosphere, or precipitate out readily. It is not uncommon for concentrations to rise for
101 | P a g e
-------
Appendix C
some distance from the emitting source, and then to fall from that peak concentration. The Gaussian plume
model applies to gases, and emissions of particulates can drop off more quickly than gases.
Releases to land may follow extremely complex patterns of dispersion. Added to that are the very site-specific
characteristics of potential human exposure via drinking water, vapor intrusion or contact with contaminated
soils, etc. For water pollution, similar complexities exist, most notably that an effluent is carried away
downstream of a running body of water, dilution can be complicated by the presence of other water entering
stream segments, by volatilization, by biological and chemical interactions, and by deposition to sediments, and
finally by the treatment and removal of a water pollutant sent to a publicly-owned treatment works.
We also note that researchers and others have taken varied approaches to representing the proximity of
facilities to populations. The EJSM model of environmental justice concerns, developed for the state of
California, scored facility proximity in concentric rings around a population centroid (Pastor Jr., Morello-Frosch,
& Sadd, 2010; Sadd, Pastor, Morello-Frosch, Scoggins, & Jesdale, 2011). All facilities within 1 mile received a
score of 3. All within the 1 to 3 mile band received a score of 2, and those between 3 and 5 miles received a
score of 1. Anything beyond 5 miles received a score of zero. This step-wise scoring represents the judgment of
the model developers, influenced by interactions with various stakeholders.
Finally, we note that EJSCREEN's measure of proximity is intended to represent more than simply real or
potential human health adverse effects coming from exposure. Some parts of the environmental justice
literature reflect semi-quantitative factors, such as increased psychological stress, fear and other reactions to
the presence of LULUs. This is not the forum for sorting through those factors.
However, we have made a judgment call: For the purposes of this EJSCREEN tool, we represent a facility's
measure of proximity by the inverse of its distance from the estimated location of the average person. A block's
proximity score is the sum of the inverse distances of all the facilities of a particular type.
Note that for the minority of block groups in the United States with no residential population, we take a
straight average of the block scores.
The units for these measures are facilities per kilometer. A block group could have a score of 1.0 if all residents
were an average of one kilometer from a single facility, and all other facilities were so distant (> 5 km) as to
make no contribution to the score. Another block group could have a score of 1.0 if there were five facilities
that were all exactly five kilometers from the residents.
Calculating Proximity to Traffic
We have adopted essentially the same approach described above for representing proximity to highway
segments - an inverse distance-weighted sum of highway segments surrounding each block, and a population-
weighted sum of the individual blocks' contributions to the block group.
The highway segment database that we have used is described in section 3. These segments differ from a
facility database in that they are lines on a geographic area, rather than points that represent the facilities. In
102 | P a g e
-------
Appendix C
our approach, we find the distance from the block centroids to the nearest part of each surrounding highway
segment. The nearest point, dij, could be an end of the highway segment or some point in between the ends.
We also multiplied each dij by the annual average daily traffic estimate that is associated with each highway
segment. This is meant to reflect the traffic intensity, and this differs from the facility approach, where we have
taken each facility within each group as having equal importance. Also, for traffic proximity, the search radius is
500 meters and the score uses distance in meters, not kilometers.
Calculating Proximity - Additional Details
We address two modifications to the general method described above. The first deals with instances where a
facility or highway segment location is very close to the centroid of the block. The second is an accommodation
to the computational intensity of the general method.
Extremely Small dij Values
Our intention is to represent the proximity of facilities or highway segments to the population within each
block. All facilities and each part of all highway segments fall within one block. By chance, some portion of
those points fall very close to the block centroids.
We do not know how the population is geographically distributed within any block, but we assume that people
are more likely to be distributed across the blocks' expanses than to be concentrated at one point, such as the
centroid. In fact for rural, suburban and many non-high rise urban areas, people's residences are more likely to
be closer to the blocks' peripheries (bounded by roads) than clustered at the centroids. Thus, when a facility
location happens to be very close to the block centroid, it would result in an artificially high contribution to the
block's score. This is not a hypothetical problem: We have observed dij values well below 100 meters, and some
below 10 meters.
In looking for solutions to the problem, we conducted analyses and arrived at the approach we have adopted.
Blocks vary widely in their total area and in their shapes. Both can be found in the Census Bureau's Tiger shape
files. Dealing explicitly with the individual block shapes would be computationally very intensive because there
are over 11 million blocks. Since we cannot easily find out how the residents are actually distributed in those
areas, we made two simplifying assumptions:
residents are evenly distributed across the surface area of each block, and
1/2
each block can be represented by a circle whose radius is [Block area / Pi] .
We call this latter value the Block Area Equivalent Radius.
Our investigations indicate that for any dij less than 90% of the Block Area Equivalent Radius, 0.9 times that
value is a reasonable representation of the average distance from the facility for all residents in the block. We
call this the dij corrected-
103 | P a g e
-------
Appendix C
Our computational scheme determines the dij values as described above, tests for the comparison with 0.9 *
Block Area Equivalent Radius, and substitutes dij corrected values. We found that we needed to make that
correction for less than 1% of all facility / block combinations in an early testing data set that used 2005-2009
ACS data.
Accommodating to Computational Intensity - Combine a Distance Limit with a Nearest Facility
Approach
Our task is to compute a proximity score for each of the facility or site types and highway segments for each of
the more than 217,000 block groups, comprised of over 11 million blocks. The number of facilities nation-wide
varies from hundreds of TSDFs to many thousands of RMP facilities. Computing all the combinations would
require more computational time and resources than were available.
In addition, doing so would be wasteful and perhaps irrelevant. The one over distance function we have chosen
to represent concerns about facilities and highways drops off greatly for most facilities beyond the nearest
ones. The miniscule contribution of a facility 100 kilometers or more from a block is not only small, compared
with those that may be within 5 to 10 kilometers, but has little common-sense meaning, in our view.
Consequently, we have followed the general approach described above only for facilities or sites within 5
kilometers of a block's centroid, and within 500 meters for highway segments. Depending on the facility or site
type, we find that 30-40% of block groups have at least one facility (RMP, TSDF, or NPDES) within the 5
kilometers limit, and almost 10% have one or more NPL sites within 5 km, in the 2015 version of EJSCREEN.
Of course, every block and block group has one nearest facility, even though it may be beyond the 5 kilometers
horizon, and some of those may be fairly close to that limit. We have also calculated the distance to the facility
nearest to each of the blocks. For those blocks lacking anything within the 5 kilometers, we represented the
facility proximity by one over the distance to that single nearest facility.
This added computational complexity to the approach, but at far less cost than computing the full matrix of
millions of blocks times thousands of facilities and sites.
This hybrid approach results in every block (and thus every block group) having a nonzero, positive proximity
score. All of the resulting block proximity scores are necessarily less than the score had we computed the full
matrix, but we judge that this is a reasonable and practical compromise. Figure 1 shows histograms for
proximity scores. Counting only the single nearest beyond 5 km has the effect of shifting scores under 0.2 to
the left, to lower scores than if all were counted, but the graphs show no major discontinuities, suggesting this
limitation (counting only the nearest one) has little impact overall.
Data and Computational Scheme
Using the Census 2010 block boundaries, the distance to all facilities within 5 kilometers of all blocks (not just
block groups) was determined, and distance to the nearest facility at any distance was determined if none were
found within 5 km.
104 | P a g e
-------
Appendix C
The dij values were compared to the 0.9 * Block Area Equivalent Radius and corrected values were used when
necessary, before computing 1 / dij. The 1 / dij values were summed for each block to compute the
BlockScorejk- These were then rolled up to the block group level, applying the population weighting described
above, for the final BlockGroupScorek .
Caveats and Observations
Several aspects of the proximity analysis approach have been mentioned above, but deserve summary here.
We recognize that our selection of the inverse of distance is a design choice that represents our
judgment of a balance among competing factors.
We recognize that one could potentially attempt to distinguish among facilities within each facility
category by quantitative or qualitative measures of importance. These could include total pounds
released or toxicity-weighted releases for NPDES facilities; the number of accidental releases
and/or their apparent severity for RMP facilities; some classification of the likelihood of releases for
NPL sites or TSDFs; and general indications of scale for all of them. We note that CalEnviroscreen
has addressed this issue to some extent, and that the RSEI tool based on TRI data may be relevant
to future work on this issue. At this point, we have chosen not to develop any such potential scaling
adjustments.
We recognize that all location data are subject to potential error. While we have high confidence in
the block centroid locations, we know that the facility or site or roadway location data may contain
larger or smaller errors, and that for large facilities or sites, one point may not be an entirely
adequate representation of the location of its releases or of neighbors' perceptions.
We recognize that the computational accommodation we describe above results in a hybrid of
measures: For some block groups, all blocks have one or more facilities within 5 kilometers and the
score is the summation of all those potentially multiple facility/block combinations; for other block
groups, none of the blocks have a facility within 5 kilometers and the score is the contribution of
the single facility closest to each block; while for some block groups, we have a mix of those
situations. We believe that this is a reasonable compromise
105 | P a g e
-------
Appendix C
Note: The method described in the above section would be the case if homes and facilities were on average uniformly
distributed within block groups that were roughly circular on average, because the average distance between two random
points in a circle of radius R is 90% of R (Weisstein, Eric W. "Disk Line Picking." From Math WorldA Wolfram Web
Resource. http://mathworld.wolfram.com/DiskLinePicking.html). This means that if a population is randomly spread over a
roughly circular block, a facility in the block typically would be 0.9R from the average person. Also, the average point in the
circle is 0.67R from the center, and 1.13R from the edge of the circle. We can describe this relationship using an equation
that is a portion of the formula for the distance between two random points in a circle of radius=l. The formula is
C f ~ \[ a + k2 - (2 k ^) cos(t) dt da
where b= the facility's distance from the center as a fraction of the radius, and the integral over a represents distances of
residences from the center. We can solve this equation using http://WolframAlpha.com, for b=0, 0.5, or 1, representing
points at the center, halfway to the edge, and at the edge of the circle. For example, we can use this equation for b=0.5 to
find that the average person, if randomly located in a circle of radius R, is a distance of about 0.8 R from a facility that is
halfway between the center and edge of the circle. For the distance between the average person and a randomly placed
facility in the circle, we use b=sqrt(0.5) instead, and the following would be used as the input to WolframAlpha:
lntegrate[(l/Pi) Sqrt[a + (Sqrt(0.5))A2 - 2 * (Sqrt(0.5)) * Sqrt[a] cos(t)], {a, 0, 1}, {t, 0, pi}] or
http://www.wolframalpha.com/input/?i=lntegrate%5B%281%2FPi%29+Sqrt%5Ba+%2B+%28Sqrt%280.5%29%29%5E2+-
+2+*+%28Sqrt%280.5%29%29+*+Sqrt%5Ba%5D+cos%28t%29%5D%2C+%7Ba%2C++0%2C++l%7D%2C++%7Bt%2C+0%2C+
pi%7D%5D+
106 | P a g e
-------
Appendix D
APPENDIX D. SUMMARY STATISTICS FOR INDICATORS
This appendix provides basic information and statistics on the environmental and demographic data used in
EJSCREEN.
Table 6 shows summary statistics for the 12 environmental indicators, including the population mean and
selected population percentiles. The mean score and high percentiles for each indicator provide useful
perspective on the magnitude of these environmental factors for the average or highly exposed individuals.
The mean neurological Hazard Index (HI) for example, is only xx, meaning that the estimated mean exposure is
only x% as high as the health-based Reference Concentration, and even the 99th percentile value is only
approximately xx. The respiratory HI, however, has a mean above xx, meaning the RfC is typically exceeded by a
factor of xx.63
ATA update pending
In the 2015 version of EJSCREEN, the PM2.5 level in the average person's block group (based on tract estimates)
was 9.7 ng/m3. Roughly 7-8% of U.S. residents had block group (based on tract) estimates above 12 ng/m3 in
these 2011 estimates. Note that 15 ng/m3 was the health-based annual ambient standard as of mid-2012, with
a revised standard of 12 ng/m3 finalized in December 2012. However, it is important to understand that the
EJSCREEN value does not indicate nonattainment of a standard because it is based on 2011 estimates of each
block group, from a combination of modeling and monitoring, while nonattainment is determined by individual
monitors (each intended to represent a relatively large area, often a county), based on three recent years of
monitoring data. Likewise, the ozone indicator cannot be compared to the ozone ambient standard.
pen
The NATA cancer risk mean people, or a lifetime individual
cancer risk of xxlO"5, which is orders of magnitude lower than typical premature mortality risk estimates
associated with recent ambient levels of PM2.5.
The facility proximity indicators generally have mean scores of 0.05 to 0.31. The NPL's mean score of 0.10 is
comparable to the average person having one NPL site 10 kilometers away. The mean RMP score of 0.31 could
result from one RMP facility at 3.2 kilometers distance, for example, and the median is roughly 0.14 (or about
1/7), meaning that most of the U.S. population has at least one RMP within about 7 km of their home. About a
third of the population lives within 5 km of an RMP facility, but less than 7% of the population has any within 1
km.
Table 7 shows a similar set of statistics for the demographic data used in EJSCREEN. The overall US percents low
income and minority were 34% and 36% respectively, and the medians were somewhat lower. The 80th
percentiles were 53% and 69%. The top 5% lived in block groups with a Demographic Index above 80%. Note it
is more common to see block groups close to 100% minority than close to 100% low income.
63 High values for the neurological HI tend to be driven by exposure to acrolein above the Reference Concentration, which
is designed to protect against the risk of nasal lesions as the health endpoint.
107 | P a g e
-------
Appendix D
Table 6. Summary Statistics for Environmental Indicators
Environmental Indicator
Missing
Minimum
25%ile
50%ile
(median)
Pop.
Mean
75%ile
80%ile
90%ile
95%ile
99%ile
PM 2.5
1664
4.3
8.8
9.9
9.8
11
11
12
12
13
Ozone
1664
25
42
46
46
51
51
54
57
66
NATA DPM
NATA cancer risk
fA update pending|
NATA respiratory HI
NATA neurological HI
% pre-1960 (lead paint)
0
0.00
0.05
0.21
0.30
0.50
0.59
0.77
0.87
0.96
Proximity Traffic
253
0.00
11
35
110
100
130
250
430
1,200
Proximity NPL
0
0.00
0.02
0.05
0.10
0.10
0.12
0.19
0.31
0.88
Proximity RMP
0
0.00
0.08
0.14
0.31
0.30
0.39
0.76
1.2
2.4
Proximity TSDF
0
0.00
0.01
0.02
0.05
0.05
0.07
0.11
0.18
0.53
Proximity NPDES
0
0.00
0.08
0.13
0.25
0.24
0.30
0.55
0.89
1.9
Source: 2015 version of EJSCREEN. See body of report for sources and definitions of environmental indicators.
Notes: Population percentiles (and means) are shown, not block group percentiles (or means), so 80%ile means 80% of the population has a lower (or
exactly tied) block group score. Values in table have been rounded to two significant digits, except for PM, ozone, and DPM, which use three significant
digits. Numbers may differ slightly from those in EJSCREEN reports. Summary statistics for a given environmental factor exclude block groups where that
environmental indicator was not available (missing).
108 | P a g e
-------
Appendix D
Table 7. Summary Statistics for Demographics
Demographic variable
Missing
Minimum
25%ile
50%ile
(median)
Population
mean
75%ile
80%ile
90%ile
95%ile
99%ile
Demographic index
0
0
17%
29%
35%
50%
57%
71%
80%
90%
% low-income
0
0
17%
30%
34%
48%
53%
65%
74%
87%
% minority
0
0
9%
26%
36%
60%
69%
89%
96%
100%
% less than high school
0
0
5%
11%
14%
21%
24%
33%
42%
60%
% linguistic isolation
0
0
0%
1%
5%
6%
8%
16%
25%
44%
% under 5
0
0
4%
6%
7%
9%
10%
12%
14%
18%
% over 64
0
0
7%
12%
13%
17%
19%
23%
28%
44%
Supplementary demog.
index
0
0
11%
16%
18%
24%
27%
34%
39%
47%
Source: 2015 version of EJSCREEN. Calculated based on 2008-2012 5-year summary file, American Community Survey (ACS), from the US Census Bureau.
Note: Population percentiles (and means) are shown, not block group percentiles (or means), so 80%ile means 80% of the population has a lower (or
exactly tied) block group value. Values in table have been rounded to an integer percentile 0-100. Numbers may differ slightly from those in EJSCREEN
reports.
109 | P a g e
-------
Appendix D
Table 8 shows Spearman correlations in block group level scores for all pairs of the 12 environmental indicators
in the 2015 version of EJSCREEN. All correlations are positive except that ozone is weakly, sometimes
negatively, correlated with the other indicators. The strongest positive correlations were among the NATA-
derived factors: cancer, neurological HI, respiratory HI, and diesel particulate matter indicators. The TSDF
proximity indicator was correlated with these four. Similarly, the traffic indicator was correlated with the NATA
factors. All other coefficients were less than 0.50.
110 | P a g e
-------
Appendix D
Table 8. Spearman Correlation Coefficients for Environmental Indicators
cancer resp. neuro. Traffic % pre- NPL RMP TSDF NPDES
PM2.5
ozone
DPM
risk
HI
HI
proximity
1960
proximity
proximity
proximity
proximity
PM2.5
1
0.23
0.22
0.22
0.21
0.30
0.34
0.18
Ozone
0.23
1
-0.03
-0.16
-0.09
0.03
-0.02
-0.06
DPM
1
cancer risk
1
|NATA update pending|
resp. HI
1
neuro. HI
1
Traffic proximity
0.22
-0.03
i
0.17
0.33
0.3
0.35
0.3
% pre-1960
0.22
-0.16
0.17
1
0.18
0.17
0.08
0.17
NPL proximity
0.21
-0.09
0.33
0.18
1
0.2
0.37
0.25
RMP proximity
0.30
0.03
0.3
0.17
0.2
1
0.25
0.35
TSDF proximity
0.34
-0.02
0.35
0.08
0.37
0.25
1
0.17
NPDES proximity
0.18
-0.06
0.3
0.17
0.25
0.35
0.17
1
Source: 2015 version of EJSCREEN. See body of report for sources of environmental indicators.
Ill | P a g e
-------
Appendix D
Table 9. Spearman Correlation Coefficients for Demographic Indicators
% less %
% % low- than high linguistic
minority income school isolation % under 5 % over 64
% minority
1
0.41
0.43
0.51
0.23
-0.34
% low-income
0.41
1
0.67
0.23
0.24
-0.15
% less than
high school
0.43
0.67
1
0.32
0.19
-0.05
% linguistic
isolation
0.51
0.23
0.32
1
0.17
-0.2
% under 5
0.23
0.24
0.19
0.17
1
-0.31
% over 64
-0.34
-0.15
-0.05
-0.2
-0.31
1
Source: 2015 version of EJSCREEN. Calculated based on 2008-2012 5-year summary file, American
Community Survey (ACS), from the US Census Bureau.
Table 9 shows the Spearman correlations in block group level scores for all pairs of the 6 demographic
factors. All correlations are positive except those between % over 64 and all other factors.
Figure 1 shows histograms or density plots of the environmental indicator data, showing the simple
distribution across block groups for each of the 12 indicators (i.e., these figures show the distribution of
block groups, not a population distribution, but the population distribution is very similar).
112 | P a g e
-------
Appendix D
Figure 1. Histograms of Block Group Environmental Indicators as ratio to mean value (log scale shows mean value as zero)
Q_
Z3
O
i_
O)
o
_o
_Q
A
1
0
PM2.5
Q_
Z5
O
i
CD
-2
~~r
-1
Ozone
Q_
Z5
O
s
O)
-2
-1
1
Air toxics cancer risk
ATA update pending
Q_
ZJ
O
s
O)
-2
~~r
-1
1
Air toxics neuro. HI
CO
Q_
=5
O
o
o
Air toxics resp. HI
V)
Q_
=5
O
J*?
o
o
Diesel PM
(/)
Q_
Z5
O
O)
O
o
CO
Q_
=5
O
O)
o
o
Lead paint (% pre-1960)
^TmnnTTTirTFIITIITrmmT^
Traffic proximity
O
O)
o
o
-2
NPL proximity
Q.
=3
O
O)
o
_o
_Q
rtltflllllllUltTtTTlTTmT^
-2 -1
RMP proximity
o
O)
o
o
-2
TSDF proximity
o
O)
o
o
r~
-2
~r
-1
NPDES proximity
Source: 2015 version of EJSCREEN. Note: Some extreme values are not shown on the x axis (for proximity indicators) and y axis (for ozone).
113 | P a g e
-------
Appendix D
[INTENTIONALLY BLANK]
114 | P a g e
-------
Appendix E
APPENDIX E. FORMULAS FOR DEMOGRAPHICS AND EJ INDEXES
The EJ indexes rely on demographic indexes combined with environmental indicators. The demographic
and EJ indexes are calculated as follows:
Demographic Index
This is the average of percent minority and percent low income in the block group. Percent low income
is defined in Appendix B, and is essentially all residents where household income is below twice the
federally defined poverty threshold, as a percentage of all those for whom this poverty ratio could be
determined (typically known for the vast majority of the block group's population).
Demographic Index = (% minority + % low-income) / 2
Supplementary Demographic Index
This is the average of percent minority, percent low income, percent less than high school education,
percent linguistically isolated, percent individuals under age 5, and percent individuals over age 64 in the
block group.
Supplementary Demographic Index =
(% minority + % low-income + % less than high school education + % linguistic isolation +
% individuals under age 5+ % individuals over age 64) / 6
115 | P a g e
-------
Appendix E
EJ Index
The EJ Index measures how much a particular place contributes to overall nationwide differences in
environmental indicator values between demographic groups. This EJ index is a combination of a block
group environmental factor, the population of the block group, and the demographic composition of the
block group. In this index, the demographic composition of the block group is the difference between
the block group's composition and the national average, as measured by the demographic index.
EJ Index =
(Environmental Indicator)
X (Demographic Index for Block Group - Demographic Index for US)
X (Block Group Population)
EJ Index with Supplementary Demographic Index
This EJ index is a combination of a block group environmental factor, the population of the block group,
and the demographic composition of the block group. In this index, the demographic composition of the
block group is the difference between the block group's composition and the national average, as
measured by the supplementary demographic index.
EJ Index with supplementary demographics =
(Environmental Indicator)
X (Supplementary Demog. Index for Block Group - Supplementary Demographic Index for US)
X (Block Group Population)
Supplementary EJ Index 1 Based on Demographic Index
This EJ index is a combination of a block group environmental factor, the population of the block group,
and the demographic index. This EJ index measures how much a particular place contributes to the total
burden faced by the subpopulations highlighted by the demographic index.
Supplementary EJ Index 1 =
(Environmental Indicator)
X (Demographic Index for Block Group)
X (Block Group Population)
116 | P a g e
-------
Appendix E
Supplementary EJ Index 1 Based on Supplementary Demographic Index
This EJ index is a combination of a block group environmental factor, the population of the block group,
and the supplementary demographic index. This EJ index measures how much a particular place
contributes to the total burden faced by the subpopulations highlighted by the supplementary
demographic index.
Supplementary EJ Index 1
with supplementary demographics =
(Environmental Indicator)
X (Supplementary Demographic Index for Block Group)
X (Block Group Population)
Supplementary EJ Index 2 Based on Demographic Index
This EJ index is a combination of a block group environmental factor and the demographic index.
Supplementary EJ Index 2 =
(Environmental Indicator)
X (Demographic Index for Block Group)
Supplementary EJ Index 2 Based on Supplementary Demographic Index
This EJ index is a combination of a block group environmental factor and the supplementary
demographic index.
Supplementary EJ Index 2
with supplementary demographics =
(Environmental Indicator)
X (Supplementary Demographic Index for Block Group)
117 | P a g e
-------
Appendix F
APPENDIX F. QUALITY CONTROL / QUALITY ASSURANCE
EPA's quality control guidelines emphasize transparency and reproducibility as useful in ensuring the
quality of data. EPA is providing a very high level of transparency in EJSCREEN by taking several steps
described here.
The EJSCREEN Technical Documentation (this document) has extensive details on the precise sources
and exact methods used, to ensure transparency. The transparency of the data inputs is also ensured
through references to further technical documentation from the providers of those data inputs, such as
the PM2.5 and ozone estimates, NATA, the Census demographic data, and the DOT traffic database.
Metadata is linked from the web-based tool, providing quick access to further technical details.
Furthermore, the full raw database of EJSCREEN indicators and indexes, and supplementary material,
will be available to expert users who wish to go beyond the web-based interface and conduct further
analysis or research. EPA also hopes to make available the Python and R code used to develop all the
indicators in EJSCREEN, including proximity scores, percentiles, and so on. Access will be provided
through the data download section of the EJSCREEN website (http://www2.epa.gov/eiscreen).
Extensive quality control/ quality assurance efforts were made in the development of EJSCREEN and a
very brief summary is provided here.
The starting point for most of the environmental indicators was information provided by EPA Offices
(i.e., latitude/ longitude data used to create proximity indicators, and the NATA results). Those sources
of information had already been subject to QA procedures in the respective offices, and the information
had already been released to the public. EPA's Office of Air and Radiation (OAR) provided PM2.5 and
ozone estimates based on public monitoring data, CMAQ results, and a fusion model to combine them.
The CMAQ and fusion model have previously been extensively documented in peer-reviewed journal
articles (Byun & Schere, 2006; Berrocal, Gelfand, & Holland, 2010a, 2010b, 2011). The lead paint
indicator was calculated from Bureau of Census data by an EPA contractor, and then independently
replicated by EPA, through separate ACS downloads and calculations. The traffic indicator was calculated
from publicly available DOT data, as explained in this report.
The calculations of environmental indicators from those inputs was conducted by an EPA contractor for
the proximity scores and the lead paint score, using their established QA/QC procedures, so EPA did not
attempt to replicate the proximity calculations. These calculations involved time-consuming proximity
calculations, and simple calculation of the lead paint indicator. The NATA and PM and ozone indicators
were simply taken directly from EPA and used in EJSCREEN.
The demographic indicators were calculated by an EPA contractor based on ACS data they obtained
from the Census Bureau. EPA was able to independently replicate 100% of the resulting indicators by
separately obtaining the raw ACS data.
118 | P a g e
-------
Appendix F
The same is true for the EJ Indexes and all of the percentiles, map bins (for color-coded maps), and
popup text fields used in EJSCREEN - the entire geodatabase was independently replicated by EPA using
only the environmental indicators as a starting point, and applying alternative algorithms and code for
development of percentiles and bins, as well as the rounding procedures defined for popup text.
EPA was also able to conduct some limited manual replication of buffer calculations, although truly
independently replicating those GIS algorithms is challenging given the need to use data on millions of
blocks and the challenge of identifying relevant blocks in a fashion that is independent of the
geoprocessing tool used for buffer analysis in EJSCREEN. Spot checks were conducted on buffer reports
to ensure raw data and percentile calculations, use of lookup tables, rounding, significant digits, and
floored percentiles were all handled correctly.
The extensive QA/QC process did uncover numerous complex data challenges early in the process, and
ultimately lead to a final database that could be fully independently replicated from environmental
indicators calculated from public information, providing strong assurance of the integrity of the data
processing and calculations.
119 | P a g e
-------
Appendix G
APPENDIX G. PEER REVIEW
EJSCREEN was submitted for peer review in early 2014, through a letter review process conducted by a
contractor with extensive experience in organizing peer review. Based on pre-defined criteria regarding
level of expertise in relevant subject areas, four experts were identified.
The reviewers were provided with a draft of this technical documentation, describing EJSCREEN's
development, purpose, and use of selected environmental indicators, demographics, and EJ indexes for
screening and mapping.
Reviewers were also provided a live webinar presentation and demonstration, along with time for
questions for EPA. One of the four was unable to attend the webinar but contacted EPA with questions
that EPA responded to in a phone conference call.
The reviews were completed in March of 2014. Each of the four reviewers provided a detailed
discussion of their technical comments, concerns, and recommendations.
All four agreed that the new environmental justice screening tool will be helpful to its users and is
generally very well done. Each did point to some weaknesses in the tool, suggesting that correcting
these shortcomings in the next version could strengthen the tool and help its users.
Of the more than 100 distinct comments from the expert peer reviewers, more than one third were
positive statements about the quality of EJSCREEN and the documentation. A sampling of direct quotes
includes the following:
"I would like to commend the EPA ... it does represent a major step forward and the EPA should
be recognized for this achievement"
"This documentation fairly represents the tremendously difficult task of creating this tool"
"very impressed by the quality of the work"
All of the reviewers also agreed that the EJSCREEN documentation is generally well-written, clear, and
easy to follow. Reviewers did ask for editorial changes, clarification, or further rationale in the
documentation, and such comments represented about one fourth of all the comments received. They
asked for clarification in some specific sections, such as more discussion of which indicators were chosen
and why, and which were left out and why. Many of these comments have already been taken into
consideration in this version of the Technical Document.
About one third of the comments were suggestions or requests for new data (in reports, maps, and data
files), typically recommending new or improved environmental indicators (e.g., air quality, water quality,
more facility types, etc.). Some comments made suggestions that would involve adding a new feature to
the EJSCREEN tool, rather than improving the data layers or the documentation. These will be taken into
consideration in discussions of possible future updates or upgrades to the tool.
120 | P a g e
-------
Appendix G
A handful of comments raised policy considerations and inherently challenging issues in screening and
mapping. These involved basic policy questions such as what is the best spatial resolution for these
maps, and whether to combine all 12 EJ indexes (Reviewers were divided on this topic). The 2015
version of EJSCREEN continues to use block groups, but recommends an emphasis on buffers as less
uncertain than a single block group estimate. It continues to use 12 separate indexes, but these issues
can be a topic of continuing discussions and exploration in the future as the public and others work with
the new tool.
On the whole, the reviewers' suggestions have already served to strengthen the tool and its
documentation, and will continue to inform discussions. By elaborating and clarifying the options and
choices made, EPA can help the users of EJSCREEN better understand its potential and its limitations.
Improved data and methods should be considered as well in future versions of EJSCREEN. EPA looks
forward to working across various offices, with stakeholders across the nation, and academics as well as
the public, on implementation and future enhancement of EJSCREEN.
121 | P a g e
-------
Appendix H
APPENDIX H. INITIAL FILTER APPROACH FOR SCREENING
What is the 80th percentile filter?
In past screening experience, EPA has found it helpful to establish a suggested Agency starting point for
the purpose of identifying geographic areas that may warrant further consideration, analysis, or
outreach. The use of an initial filter promotes consistency and provides a pragmatic first step for EPA
programs and regions when interpreting screening results. For early applications of EJSCREEN, EPA
identified the 80th percentile filter as that initial starting point. In other words, an area with any of the
12 EJ indexes at or above the 80th percentile nationally should be considered as a potential candidate
for further review. Further review may include considering other factors and other sources of
information such as health based information, local knowledge, proximity and exposure to
environmental hazards, susceptible populations, unique exposure pathways, and other federal, regional,
state, and local data. This filter is simply a starting point, and program offices and regions should
perform additional analysis before making any decisions about potential environmental justice issues. As
EPA gains further experience and insight into the performance of the tool and its applicability for
different uses, program offices and regions may opt to designate starting points that are more inclusive
or specifically tailored to meet programmatic needs more effectively.
The 80th percentile filter in EJSCREEN is not intended to designate an area as an "EJ community."
EJSCREEN provides screening level indicators, not a determination of the existence or absence of EJ
concerns. Nor does the use of the 80th percentile filter suggest that all of the 12 environmental
indicators are equal in terms of their impact on human health and the environment. Instead, the 80th
percentile filter encourages programs to consider environmental indicators outside of their areas of
concentration. The Agency may revise this approach in the future based on experience. This 80th
percentile filter is for internal EPA use and is not intended to apply to States or other organizations.
122 | P a g e
-------
Appendix H
[page intentionally blank]
123 | P a g e
------- |